Color Space and Its Divisions

Color Space and

Its Divisions
Color Order from
Antiquity to the Present

Rolf G. Kuehni

A John Wiley & Sons Publication

Copyright © 2003 by John Wiley & Sons, Inc. All rights reserved.

Published by John Wiley & Sons, Inc., Hoboken, New Jersey.

Published simultaneously in Canada.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in

any form or by any means, electronic, mechanical, photocopying, recording, scanning or
otherwise, except as permitted under Sections 107 or 108 of the 1976 United States Copyright
Act, without either the prior written permission of the Publisher, or authorization through
payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400, fax 978-750-4470, or on the web at
www.copyright.com. Requests to the Publisher for permission should be addressed to the
Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030,
(201) 748-6011, fax (201) 748-6008, e-mail: [email protected].

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best
efforts in preparing this book, they make no representations or warranties with respect to the
accuracy or completeness of the contents of this book and specifically disclaim any implied
warranties of merchantability or fitness for a particular purpose. No warranty may be created
or extended by sales representatives or written sales materials. The advice and strategies
contained herein may not be suitable for your situation. You should consult with a professional
where appropriate. Neither the publisher nor author shall be liable for any loss of profit or any
other commercial damages, including but not limited to special, incidental, consequential, or
other damages.

For general information on our other products and services please contact our Customer
Care Department within the U.S. at 877-762-2974, outside the U.S. at 317-572-3993 or
fax 317-572-4002.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in
print, however, may not be available in electronic format.

Library of Congress Cataloging-in-Publication Data

Kuehni, Rolf G.
Color space and its divisions : color order from antiquity to the present / Rolf G. Kuehni.
p. cm.
Includes index.
ISBN 0-471-32670-4 (cloth : acid-free paper)
1. Color vision. 2. Colors. I. Title.
QP483 .K84 2003
152.14¢5—dc21 2002014045

Printed in the United States of America

10 9 8 7 6 5 4 3 2 1
 To the memory of Dorothy Nickerson and David L. MacAdam, and to
Andreas Brockes for their encouragement to continue the pursuit of color
order and color difference

All that is alive tends toward color, individuality, specificity, effectiveness,

and opacity; all that is done with life inclines toward knowledge, abstraction,
generality, transfiguration, and transparency.
Johann Wolfgang von Goethe
 Contents

Preface xii

Chapter 1. The Concept of Color Space and Color Solid 1

1.1 Introduction / 1
1.2 Divisions of Color Spaces and Solids / 6
1.3 Uniform and Regular Color Spaces / 8
1.4 Color Space, Sensation, Perception, and Awareness / 15
1.5 Plan of the Book / 16

Chapter 2. Historical Development of Color Order Systems 19

2.1 Color and Color Order Systems / 19
2.2 From Ancient Greece to the Middle Ages / 21
2.3 Color Order in the Renaissance / 32
2.4 Newton’s Color Diagram / 43
2.5 Development of the Color Circle / 46
2.6 Mayer and Lambert’s Color Solids / 51
2.7 Color Circles from Harris to Henry / 55
2.8 Three Primary Color Theories / 59
2.9 Runge’s Color Sphere / 59
2.10 The Cylindrical System of Matthias Klotz / 63
2.11 The Early Development of Psychophysics / 63
2.12 Chevreul’s Hemispheric System / 66
2.13 Doppler’s Sphere Octant / 69
2.14 Yellow, Red and Blue, For a Time Firmly Established as
Primary Colors / 69

vii
viii CONTENTS

2.15 Helmholtz, Grassmann, and Maxwell / 72

2.16 Hering / 75
2.17 Geometrical Systems of the Nineteenth Century / 78
2.18 The Nineteenth-Century Experimental Psychologists / 82
2.19 The Munsell System / 84
2.20 Ridgeway’s Color Atlas / 86
2.21 Ostwald’s Farbkörper (Color Solid) / 89
2.22 Geometrical Systems of the Twentieth Century / 90
2.23 Rösch-MacAdam Color Solid / 91
2.24 The Luther-Nyberg Color Solid / 93
2.25 The German DIN6164 System / 93
2.26 Optical Society of America Uniform Color Scales / 95
2.27 Swedish Natural Color System / 96
2.28 Universal Color Language / 98
2.29 Color Mixing Spaces / 100
2.30 Spectral Spaces / 101

Chapter 3. Psychophysics 104

3.1 Fundaments of Psychophysics / 105
3.2 Categories / 108
3.3 Differences versus Magnitudes / 109
3.4 Psychophysical Scaling: Levels of Measurement / 113
3.5 Scaling Methods / 116
3.6 Unidimensional Scaling Methods / 117
3.7 Psychometric Function / 119
3.8 Multidimensional Scaling / 119
3.9 Psychological and Psychophysical Spaces / 121
3.10 Psychophysical Scaling as a Basis of Color Space / 122

Chapter 4. Color Attributes and Perceptual Attribute Scaling 123

4.1 Theories of Vision / 123
4.2 Historical Development of Views on Attributes / 124
4.3 Whiteness and Blackness / 129
4.4 Evans’s Five Color Attributes / 131
4.5 Common Color Attribute Definitions / 132
4.6 Confirmation of Three Attributes / 134
4.7 Contrast versus Similitude / 137
4.8 Neural Correlates of Color Attributes / 137
4.9 Psychological (Perceptual) Scaling of Color Attributes / 138
4.10 Perception of Color Differences / 154
 CONTENTS ix

Chapter 5. Psychophysical Scaling of Color Attributes:

Stimulus and Perception 157
5.1 Requirements for a Uniform Psychophysical Color Space / 157
5.2 Postulated Relationship between Psychological and Physical
Magnitudes / 158
5.3 Photometry and Brightness/Lightness / 159
5.4 The Colorimetric System / 161
5.5 Cone Response Space / 164
5.6 Opponent Color Space / 169
5.7 How Are the L, M, S and X, Y, Z Color Spaces Related? / 174
5.8 Expressing Psychological Scales in Psychophysical Spaces / 176
5.9 Color Matching and Appearance Scaling / 193
5.10 Placement of the Red and Green Unique Hues in the
Opponent Color Diagram / 194
5.11 Curvature of Lines of Constant Hue Blue Colors / 196
5.12 Munsell Colors in the L, M, S and X, Y, Z Spaces and
the a, b Diagram / 196
5.13 Suprathreshold Small Color Differences / 199
5.14 Difference Threshold Measurements / 201
5.15 How Many Colors Can We Distinguish? / 202

Chapter 6. Historical Development of Color Space and Color

Difference Formulas 204
6.1 Line Elements / 204
6.2 Projective Transformations / 208
6.3 Fitting Models to the Munsell System / 213
6.4 Judd’s Model of Müller’s Theory of Color Vision / 215
6.5 Color Difference Thresholds and Matching Error / 216
6.6 Further Development of Formulas Based on Opponent
Color Systems / 224
6.7 New Small Color Difference Data / 225
6.8 Ellipse and Ellipsoid Fitting / 226
6.9 Controversies of Detail / 227
6.10 Dependence of Calculated Color Difference on
Metric Chroma / 228
6.11 The CIE 1976 L*a*b* and L*u*v* Spaces / 229
6.12 Friele’s FCM Formula / 232
6.13 Richter’s LABHNU2 Formula / 234
6.14 Weighting of Metric Lightness, Chroma, and Hue
Differences / 234
6.15 New Sets of Visual Data / 236
6.16 New Formulas / 239
x CONTENTS

6.17 Color Space Formulas and Comprehensive Models of

Color Vision / 248
6.18 Is the Opponent Color System “Soft Wired”? / 263
6.19 Spectral Spaces / 264
6.20 Performance Comparison of Various Formulas / 264

Chapter 7. Major Color Order Systems and Their

Psychophysical Structure 271
7.1 The Munsell Color System / 272
7.2 Optical Society of America Uniform Color Scales
(OSA-UCS) / 287
7.3 The Swedish Natural Color System (NCS) / 301
7.4 The “Fragility” of Color Atlases / 309

Chapter 8. From Color-Matching Error to Large Color Differences 311

8.1 A Common Basis for Comparison / 311
8.2 Chromatic and Lightness Crispening Effects / 313
8.3 Chromatic Crispening Fades as a Function of the Size of the
Difference / 318
8.4 Size and Ratio of Unit Increments / 320
8.5 Direction of Unit Chromatic Contours in the L, M, S and X, Y,
Z Spaces / 322
8.6 The Paradox of Hue Differences / 325
8.7 Unit Difference Contours around the Hue Circle / 328
8.8 Global Differences / 332
8.9 How Fundamental Are the Various Kinds of Data? / 332

Chapter 9. Conclusions and Outlook 336

9.1 What Are Color Spaces and How Can They Be Justified? / 337
9.2 What Causes the Perception of Colors and Their
Differences? / 339
9.3 Why Is Our Basic Color Experience Three-Dimensional and
Why Are There Four Unique Hues? / 340
9.4 What Are the Fundamental Perceptual Color Attributes? / 342
9.5 How Are Hue, Chroma, and Lightness Perceptions
Combined? / 343
9.6 What Causes the Perception of the Magnitude of Color
Differences? / 344
9.7 Chrispening Effects / 345
9.8 Perceptual Increment Magnitude as a Function of Stimulus
Increment Magnitude / 346
9.9 How Well Do Formulas Predict Perceived Color
Differences? / 347
 CONTENTS xi

9.10 Is Uniform Color Space Euclidean? / 348

9.11 Unique Hues and Uniform Color Space / 350
9.12 Evidence for the Operation of an Opponent Color System / 350
9.13 Opponent Signals: The Source of Hue and
Chroma Perceptions? / 351
9.14 The Approximate Shape of a Uniform Color Solid / 352
9.15 A Research Program / 353
9.16 Kinds of Spaces / 357

Notes 361

Glossary 366

References 375

Credits 399

Index 403
 Preface

Our color experiences are an important component of our visual experiences

and as such form a significant aspect of our consciousness. Color experiences
are the outcome of processing by the brain of information acquired as a result
of interaction of light energy with the three types of retinal cone cells in the
eyes. The three cone types filter the spectral complexity of our surroundings
and from the three kinds of signals the brain constructs complex experiences
that allow us to interact with our surroundings in a purposeful way. Luminance
contours are important input for the generation of these experiences but
much of the original complexity of the information arriving at the eye is in
terms of spectral signatures translated by the brain into colors. It has been
estimated that we can distinguish between color experiences that number in
the millions.
Given the human predilection for ordering experiences and given the large
number of different possible color experiences, it is not surprising that the
question of how to bring order to this multiplicity is one that humans thought
about since antiquity. Confusion created by the seemingly different result of
mixing colored lights and colored materials has complicated the search for
answers considerably. Only in the later eighteenth century have color experi-
ences begun to be sorted into three-dimensional arrangements, and three
attributes of color experiences have only been defined unambiguously in mid-
nineteenth century. Ordering color into a uniform color space means the cre-
ation of a geometrical model that is considered to be isomorphic (one-to-one
correspondence) with experiences. It is evident that this cannot be a simple
effort. Alternately, it can mean ordering color stimuli in a regular way so that
the most general psychological ordering principles are observed.
xii
 PREFACE xiii

This text presents a history of the significant steps in development of think-

ing about color order in the Western world from ancient Greece to the present.
Not surprisingly, given the complexity of the matter and the fact that it involves
a sense, several fundamental questions continue to be unanswered. Among
other things this has to do, despite hundreds of years of concentrated effort
by many individuals, with the absence of a scientifically satisfactory experi-
mental database of how humans perceive colors and color differences. The text
is limited to issues of color space as viewed against a simple achromatic sur-
round. It does not address issues of color appearance under widely varying
surround conditions.
Every space is divisible and a given color space is inextricably linked with
the definition of the divisions used. A dividing line is drawn between uniform
color spaces where distances in all directions are isomorphic to perceived dif-
ferences of equal magnitude and general color spaces that are ordered accord-
ing to some other principles. The plural is used because it has become evident
that there is no single uniform color space but each such space is related to
a quite highly specific set of viewing and general experimental conditions.
Uniform color spaces are of particular interest for color quality control pur-
poses in industries manufacturing colorants or colored goods. A considerable
variety of general regular spaces are in use, for example, in the graphics indus-
try and in computer display technology.
Regular arrangements of colors fit into euclidean space and can have many
different simple geometrical forms, depending on the definition of distance. It
is not evident that a uniform color space can have euclidean form (there are
no obvious reasons why it should).
The book begins with a general introduction to the subject in Chapter 1.
Following the historical account of color order systems in Chapter 2, funda-
mentals of psychophysics, the branch of science concerned with the relation-
ship between stimuli and experiences, are presented in Chapter 3. Chapter 4
describes the results of perceptual scaling of colors according to attributes. In
Chapter 5 these scales are related to scales based on psychophysical modifi-
cation of physical measurements (reflectance or spectral power distribution
measurements). The history until the present of the development of mathe-
matical color space and difference formulas is described in Chapter 6. Three
of the color order systems presented in Chapter 2 have been selected for more
detailed description of their development, their psychophysical structure, and
the problems associated with them in Chapter 7. Chapter 8 contains an analy-
sis of the agreements and discrepancies in psychophysical data describing
color at levels of difference ranging from color matching error to large.
Chapter 9, finally, draws conclusions and offers an experimental plan for the
kind of reliable, replicated perceptual data needed to make progress in this
field.
Aside from offering the first extended historical account of this fascinating
field, the book contains new analytical results of perceptual and psychophys-
ical color data and a synthesis of data developed for different purposes and
xiv PREFACE

under different circumstances. I believe it to be not only of interest to experts

and educators in industry and academe but also to neuroscientists and philoso-
phers grappling with problems of awareness and consciousness, to designers,
graphic artists, art historians, students of vision, psychology, design, and, last
but not least, the general reader with interest in the subject matter.
This book represents the culmination of some forty years of interest in the
subject. During these years I had many discussions with fellow devotees in
industry, academe, the Inter-Society Color Council, and other organizations
that helped shape my knowledge and views. I am grateful to all that helped
me to see the issues clearer.
The text was read and commented on in its entirety by Dale Purves. Indi-
vidual chapters have been read by Larry Hardin and Andreas Schwarz. I am
grateful to all three, but any remaining errors are my own.
A note about certain conventions: An author’s name with an associated
year relates to a complete reference at the back of the book. Comments in
angled parentheses are by the author, except in Chapter 2 where they usually
contain Greek and Latin color names.
Fig. 1-1 Images of chips of the OSA-UCS system. Left: Color 000; center: color 00-4; right:
color 00-8.

Fig. 2-12 C.B.’s hand-painted color circles, 1708. Left: The seven-color circle; right: the
twelve-color circle. Note that the pigments used for some of the colors have deteriorated.

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.
Fig. 2-13 Continuous color circle of Ignaz Schiffermüller, with twelve classes of colors, 1771.
Fig. 2-14 Schiffermüller’s tonal scales of three blues, from his 1771 work. Some of the col-
orations have deteriorated.
Fig. 2-18 Depiction of Lambert’s color pyramid, 1772. The lowest level contains the 45 colors
identified in Fig. 2-17. The higher levels contain reduced sets at higher lightness, ending in
white on top of the pyramid. Black is located on the lowest level. The colors displayed on the
front of the model represent well-known artist’s pigments of the time.
Fig. 2-19 Prismatic version of Moses Harris’s color circle of 1786. Some deterioration of col-
orants is evident.
Fig. 2-22 Hand-colored copperplate of Runge’s Farben-Kugel (1810). Views toward the white
and black poles are on top. The equatorial cross section is on bottom left and the polar cross
section on the right. There are four saturation steps between the full color on the surface and
the middle gray in the center of the sphere.
Fig. 2-25 One of twelve chromatic scales by Chevreul, with local discoloration. The colors
range from white through the full color (grade 11) to black. These scales are located on the
base plane of the hemisphere.
Fig. 2-32 Hering’s diagram illustrating the composition of mixed hue perceptions from the
unique hues located on the main axes. The fractions of blue and red of three mixed hues are
shown bottom right, 1905–1911.

Fig. 2-34 View toward the top, white corner of Benson’s tilted color cube, 1868. The gray
scale is hidden behind the white sphere.
Fig. 2-41 Artist’s rendition of Munsell’s balanced color sphere, patented in 1900. The sphere
was rotatable to achieve additive color mixture to gray and thereby show the “balance” of the
colors on the sphere. The mirror in the back discloses the blue region of the sphere.
Fig. 2-42 Constant hue page from a modern version of the Munsell Book of Color. The gray
scale is not shown. The chroma scale begins at 1 and continues from 2 at two-grade intervals
to chroma 14. Value grades are shown from 2.5 to 9. Courtesy Gretag-Macbeth Company.
Fig. 2-43 Page from Ridgway’s color atlas showing three reddish blue hues lightened from
the central color in three steps toward white and darkened in four steps toward black, 1912.
Fig. 2-45 Vertical cross section through Ostwald’s double-cone color solid illustrating constant hue colors 1 and 13, with veiling toward white
and black. The achromatic scale is at the center. From Farbkörper, undated.
Fig. 2-51 View of MacAdam’s model of the OSA-UCS color solid illustrating the existence of
several cleavage planes. See text for more detail. Slide courtesy D. L. MacAdam.
Fig. 2-53 NCS constant hue triangle of hue Y90R with full color C, white W, and black
S. Colors of constant blackness s are located on lines parallel to W–C, colors of constant
chromaticness c fall on lines parallel to W–S. Courtesy NCS.
Fig. 2-54 Color Picker screen from Adobe® Photoshop showing specification of a given reddish blue
color in four systems: HSB, Lab (=L*, a*, b*), RGB, and CMYK.
Fig. 7-16 Image of the 43 hexagonal enamel color plates used by the OSA-UCS committee
to establish the fundamental perceptual data for the system. Note that there is no achromatic
color among them.
 Chapter 1
The Concept of Color
Space and Color Solid

1.1 INTRODUCTION

Attempting to understand our place in the world and classifying things and
experiences is a well-known human trait. Already ancient Greek philosophers
thought about the multitude of color perceptions, but they despaired of finding
a system in which to place them. First, colors were logically sorted according
to lightness, regardless of hue. Early in the second millennium we begin to find
descriptions of tonal scales of individual hues or mixed tones, like flesh color.
They were achieved by adding lighter or darker pigments of similar hue, even
black or white, to saturated chromatic pigments. Systematic hue circles began
to appear in the late seventeenth century. The concept of a three-dimensional
logical arrangement of color perceptions began to take shape only in the eigh-
teenth century.
Color space is a three-dimensional geometric space with axes appropriately
defined so that symbols for all possible color perceptions of humans or other
animals fit into it in an order corresponding to the psychological order. In this
space each color perception is represented as a point. The symbolic represen-
tations of color perceptions in this space form the color solid. The earliest pro-
posals for color solids had simple geometrical forms: triangular double
pyramid, sphere, cone, and so forth. There is, of course, no a priori reason why
a systematic arrangement of color perceptions should fit into a simple geo-
metrical solid. What controls the form of the solid is the definition of the axes
of the space and their divisions.

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

1
2 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

There is ample evidence that the colors we experience in various conditions

from a given spectral stimulus can vary widely. According to one view they are
determined by empirical rules derived on an evolutionary basis for our species
and for each individual. There is strong evidence that the color attributed to
an object depends on the nature and complexity of the surround in which the
object is seen. In scientific experiments the complexity of surrounds usually is
minimized (elementaristic approach).1 Color experiences from given stimuli
under elementaristic conditions depend on the exact conditions and change to
a smaller or larger extent as quality and complexity of the surround and light-
ing change. Only under closely controlled conditions can a color space for the
average color normal human observer be represented by spectral stimuli. In
these relativized circumstances terms such as color stimulus and object color
have applicability restricted to the experimental conditions and cannot claim
the level of universality that has generally been assumed from the eighteenth
to the twentieth century. Critics of the idea of color space have pointed to its
lack of solid foundation. While this is ultimately true in the end such criticism
appears simply to address the fact that at this point in time we do not have an
understanding of consciousness. Color perceptions are the result of brain activ-
ity; they are subjective and private. As for all other sensory feelings and beliefs
we do not know how in a given situation a given light stimulus can result in
our seeing an object, and this object to have the appearance of red. It is not
clear that humans will ever gain an understanding of this process. Color sci-
entists have over the years built conjectural models based on what must, in
the absence of true knowledge, be called coincidental relationships between
stimuli as viewed in controlled circumstances and visual perceptions. In a
perfect world this is not an acceptable process. Given the lack of fundamen-
tal understanding of consciousness it is an empirical approach having
produced many reasonably well established, coincidental or otherwise, rela-
tionships.
Within the framework of an evolutionary development model, some key
questions concern what forces in our early history shaped the development of
visual sense and what strategies were implemented during its evolution by
neurochemistry to deal successfully with the pressures of these forces. The
simplistic color perceptions and attributes on which color scaling is based are
doubted by some psychologists as having anything to do with the fundamen-
tal perceptual processes embedded in our visual system as a result of interac-
tions with the environment. We appear to be only at the beginning of a process
to find answers. Questions such as why color space is (at least) three-dimen-
sional and why there are four psychologically fundamental hues and not more
or less have started to be asked only recently.
The issue of a systematic arrangement of color perceptions under simpli-
fied viewing conditions is a relatively abstract matter, removed from such con-
siderations. It is probably not surprising that it took shape in the age of
Enlightenment with its belief in a universal rational order. In the twentieth
century, aside from fundamental considerations of trying to understand our
 INTRODUCTION 3

place in the world, the quest was shaped by technical and economic issues of
color control of manufactured colored goods.
A color space belongs in the domain of psychology. The description of
stimuli that under standard conditions result in perception of colors in that
space is an aspect of physics. Together they form the uneasy domain of psy-
chophysics that attempts to connect stimuli with perceptions (see Chapter 3).
The stimuli are messages to us from the outside world. An alternative view is
that we actively search for them when viewing the world. They enter through
the pupils of the eyes and are absorbed by the retinal layer. There they trigger
a complex chain of events that result in our perceptions. These events
belong into the domain of neuroscience and are part of the conundrum of
consciousness.
The number of different color experiences we can have is unknown, but
large. Given a particular starting point in color space the finest perceptual divi-
sion of color space is represented by visual threshold increments deviating
from that point in all directions. A color space of given definition can only be
expressed in terms of differences within the related color solid against a
chosen surround because it is only applicable to those conditions. The small-
est difference in a color solid as related to a given starting point, therefore,
consists of a pair of different color stimuli displayed against a particular
(usually neutral) surround and seen as having a just perceptible difference.
Generally, a color space and the related color solid may be defined as an
economic systematic description of subjective color experiences, and as such
it is not subject to engineering precision. It is indicative of our visual strate-
gies vis-à-vis the world.

Personal Color Spaces and Color Solids

Each person with normal color vision has individual, personal (relativized)
color spaces and related color solids (depending on the conditions under which
they were established). Such individual solids vary within limits, based on the
detailed implementation in an individual of his/her color vision apparatus.
(Relativized personal spaces generally are at least in ordinal if not in interval
order compared to that of the average observer.)2 What it means is that if the
reader and the writer sense the spectral power distribution representing a par-
ticular object color field in a particular surround and illumination, the result-
ing experience is likely to be somewhat different. Such a statement assumes
that both observers are “color normal” and that color normal individuals have
in essence the same fundamental color experiences. It does not consider the
possibility, raised by some philosophers, of what is loosely called “spectrum
inversion.” It cannot be excluded with certainty that, for example, the reader
actually experiences as green what the writer experiences as red, regardless of
how it is named.
How different the experience resulting from a given spectral power distri-
bution might be in terms of hue can be judged from individual determination
4 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

of unique hues and, to less extent, from color perceptions judged to be inter-
mediate between unique hues. Unique hues are those four primary hues that
do not contain perceptual components of other hues. A unique red hue is
neither yellowish nor bluish: it is just red. Color stimuli resulting in unique hue
perception vary among color normal observers.3 This variation depends on the
hue in question. It ranges approximately from 5% to 12% of the total hue vari-
ation in a hue circle experienced under standard viewing conditions (i.e.,
approximately two to five Munsell 40-hue steps; see Chapters 2 and 7 for infor-
mation on the Munsell system). Because of the absence of unambiguous cri-
teria, it is not possible to meaningfully assess the stimulus variability for other
hues. It is quite evident that there is also variability in the experience of gray
scale steps, in adaptation and constancy response and other visual mechanisms,
resulting in considerable variability of individual experience when looking at
a given scene of color stimuli. Persons with impaired color vision have implicit
color spaces significantly different from those of color normal observers.
Their nature cannot be conveyed with certainty. Theoretical considerations
of the genetics of color vision indicate that as much as 50% of the female pop-
ulation have the potential for four rather than the normal three cone types
even though none has so far been identified as having four cone types.4 Richer
color experiences than those had by standard trichromatic observers have
recently been determined for females with the genetic potential for four cone
types. In how many ways their color experiences are richer remains to be
determined.

Adaptation and Conspicuousness of Differences

Color experiences, in the normal case, result from the impact of light energy
on the retina in our eyes. They are known to depend on the absolute level
of light energy. This level can differ by a ratio of 1 million to 1 (on a retinal
illumination basis). There are mechanical (pupil size) and neurochemical
processes to manage such large variation, known under the general term of
adaptation.5 The complete process adjusts the range of incoming light to an
output capability with a range of approximately 100 : 1. The adaptation process
works to map the energy pattern in a given viewing situation to the total
output range so that contrasts between different areas are seen roughly as the
same under a wide range of illumination. At very low intensities of light we
see no hued colors and neither do we at very high intensities. Probably because
of the importance of very low light levels (night) in the lives of some early
ancestors, we have a separate set of receptors for that situation, the rods. Rod
signals pass through the same postreceptoral cells into the brain as cone
signals. If they have any effect on daylight color vision, it is very small. The
response pattern as a function of light intensity of our daylight-level sensors,
the cones, is S-shaped, but the response has a considerable range that is
approximately linear in the center region. There are issues at low and high
levels of response of cones that cannot be of concern in this discussion. This
 INTRODUCTION 5

text is largely limited to color spaces and solids represented by reflecting

materials at mid levels of illumination, say 500 to 1500 lux.6
Earlier, mention was made of the lack of constancy of color perceptions
resulting from most stimuli as a function of surround or illumination changes.
Chromatic adaptation is a seemingly opposite process. Its purpose appears to
be to provide a considerable level of color constancy for reflecting objects with
certain spectral signatures. Among terrestrial nonhuman mammals trichro-
matic color vision is limited largely to fruit eaters and pollinators. For them it
is important to recognize their objects of interest in all natural lighting condi-
tions. Color is an important part of the stored memory of the appearance of
objects and helps to recognize them rapidly when encountered again. Without
chromatic adaptation, colored objects in the natural world might change their
appearance significantly over time as a result of changes in ambient illumina-
tion. With independent adaptation capabilities for each cone type, likely
together with additional processes, our ancestors could recognize the colors of
most natural objects as essentially the same regardless of the quality of illu-
mination and surround. This is less true today than it was at a time when there
were only natural objects and all light was sunlight, direct, scattered, or
reflected. Today we have a large number of artificial colorants and various arti-
ficial light sources that have complicated the issue considerably, resulting in
smaller or larger changes in the appearance of objects as a function of sur-
round and illuminant. This text does not consider most issues of chromatic
adaptation but considers color spaces and related solids only in terms of
colored objects as viewed against achromatic backgrounds of varying levels of
lightness under a standard light source.
The visual system has developed in a way that favors the conspicuousness
of small differences in reflectance signatures of objects. Its cause may have
been an escalating battle between camouflage and detection, a matter of life
and death. Highest discrimination of small reflectance differences between
objects is provided in a surround with reflectance intermediate to those of the
objects compared. This results in improved detection of highly camouflaged
predators or prey in natural surroundings. The principle also applies when the
number of objects with different color increases and/or the differences
between them become larger. Best discrimination is provided in this case by
an average (i.e., mid-level) achromatic surround. The best surround to view a
complete color atlas, by this reasoning, is a mid-level gray.

Mathematical Color Appearance Models

Our complete color experience is much wider than what was just discussed.
We view natural scenes, color television, computer monitors, projected slides,
projected digital images, the output of many coloration devices under many
different light sources, metameric objects under different light sources, and so
on. It has become important to be able to predict for an average observer
the appearance of colored objects in many different conditions. This is the
6 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

province of color appearance modeling, evolving rapidly in the last ten years
and having developed several mathematical models that are still compara-
tively simple and correspondingly only modestly accurate. This is expressed to
some extent by the fact that several different modeling approaches can result
in about the same level of prediction accuracy. An excellent survey of color
appearance modeling has recently been provided by MD. Fairchild (1998). As
indicated, the present text is concerned with color appearance under limited
conditions only.

1.2 DIVISIONS OF COLOR SPACES AND SOLIDS

Color spaces and solids are always expressed in terms of differences of some
kind between color perceptions. As will be shown, there are various kinds of
differences that have been proposed for color spaces. A kind of color space of
particular interest is one in which distances in the solid in all directions are
proportional to the magnitude of perceived differences between the related
color experiences. Such a space can be built from (or divided into) threshold
differences or larger differences. Differences imply scales, and there are
several different kinds of scales possible. The primary scales are psychological
or perceptual. Such scales are built on the basis of perceptual attributes. A
logical expectation is that the perceptual attributes form the axes of the space.
For simple observation situations (uniform achromatic surround and defined
light source) three attributes are sufficient to define the perceived color of an
object. If its dependence on surround and illumination is to be considered
quantitatively, additional attributes are required (see Chapter 4). We will find,
however, that all possible hue perceptions are best ordered in a circle and that
the hue attribute, therefore, is a function of two dimensions of the space.
A psychological color solid and the space into which it fits can be built from
a very large number of color samples. It requires picking the appropriate
samples to represent the chosen type and size of difference. Once the task is
complete, the selected samples represent the solid and the space. This is not a
generally satisfactory solution because producing many copies of the solid
requires large sheets of uniformly colored materials. Our inability to define
color experiences from an object verbally or by some other subjective means
with a high degree of accuracy and precision makes it desirable to use objec-
tive means of defining the color samples. Weight of colorants in a mixture has
been used in the earliest attempts at illustrating color scales, (e.g., F. Glisson,
Chapter 2). With the development of photometry in the eighteenth century
and colorimetry in the late nineteenth century, physical and psychophysical
means of specifying color stimuli became available. It quickly was learned that
in a given set of conditions the relationship between measured stimuli and per-
ceptions is not linear, and the next task was to develop models of the rela-
tionship between physical properties of stimuli and resulting perceived color.
This is the domain of psychophysics. This branch of psychology developed
 DIVISIONS OF COLOR SPACES AND SOLIDS 7

as a quantitative science in the mid-nineteenth century. Both psychological

color scaling and psychophysical color modeling continue to be incomplete
activities.
The simplest kind of scale applicable to the universe of color experiences
is the ordinal scale. It describes the order of entities that form the scale. An
example of an ordinal scale is a random series of gray papers, arranged in
terms of perceived lightness. The differences between steps are also random
in perceived size. A psychophysical example is a gray scale in which the steps
differ by 10% in luminous reflectance. It is not a perceptually uniform scale.
There are an infinite number of possible color spaces based on such ordinal
scales and encompassing three dimensions. Examples are cone sensitivity
spaces, the CIE tristimulus X, Y, Z and x, y, Y spaces, the Luther-Nyberg space,
or spaces based on color matching functions different from the standard CIE
functions (see Chapter 5).7
Of historically greater interest have been interval scales of object color per-
ceptions and the psychological space derived from them. Interval scales, as
defined in Chapter 3, are scales where the meaning of the size of the interval
is the same regardless of where on the scale the interval is located. Interval
scales are known as psychometric scales. Typical interval scales are lightness,
hue, and chroma scales—or scales of complex color differences—based on per-
ceived total color difference. An example of approximation of a uniform psy-
chological color solid based on interval scales is the Munsell “tree” of colors,
a three-dimensional logical arrangement of color chips forming interval scales
in certain directions. However, equal geometrical distances in the Munsell
system do not correspond in all directions to equal perceived differences.
A different approach has been pursued in the development of the Optical
Society of America Uniform Color Scales (OSA-UCS; see Chapters 2 and 7).
Here no attributes have been scaled, but uniformity in size of complex chro-
matic color differences in a triangular grid was determined at approximately
constant lightness, as was the magnitude of combined chromatic and lightness
differences. A fitted formula was then used to tile the corresponding space so
that colors in twelve directions were defined from a central color approximately
perceptually equally distant. The result is a space with a square grid pattern for
colors of equal perceptual lightness rather than the radial grid of the Munsell
system. It turns out that also here a uniform solid has not been formed. Uni-
formity of color space and the related color solid has been a goal since the ear-
liest attempts at building color appearance spaces. Its importance increased
with the capabilities of accurately and inexpensively measuring reflectance
properties and the related opportunities for objective color quality control in
the mid-twentieth century. This text, while discussing many different kinds of
color spaces, pays particular attention to the issues of uniform color spaces.
Ratio scales represent the next higher level of scale complexity. Here not
only are the intervals quantitatively the same, but ratios are also fully valid.
Historically ratio scales for colors have been controversial. Many observers do
not agree that it is possible to make a judgment that a given color is twice as
8 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

Fig. 1-1 Images of chips of the OSA-UCS system. Left: Color 000; center: color 00-4; right:
color 00-8. (See color plate.)

red or twice as black as another color. The difficulties involved can be visual-
ized by comparing, say, OSA-UCS chips 000, 00-4, and 00-8 (Fig. 1-1). If the
OSA-UCS greenness–redness scale could be considered a ratio scale, the
statement describing the g-8 chip as twice as red as the g-4 chip should apply.
But many observers, including the author, are not prepared to agree with such
a statement.
Once a color solid has been perceptually developed for a specific set of con-
ditions, the selected stimuli/samples can be defined physically by spectral
power or reflectance measurements. The next step is to build a mathematical
model connecting the physical with the psychological data in a manner result-
ing in perfect or near perfect agreement between the two sets. As will be seen
in Chapter 6, much effort has been devoted to finding the mathematical defi-
nition of a uniform psychophysical color space. There are a number of prob-
lems and difficulties with such efforts. They begin with the difficulties or
impossibility of creating an euclidean geometrical model of a uniform psy-
chological color space. In addition the physical definition of samples and spec-
tral power distributions, representative of the observed objects, is not without
problems.

1.3 UNIFORM AND REGULAR COLOR SPACES

The Oxford English Dictionary’s definition of uniform in regard to motion or

dimensions is “free from fluctuation or variation in respect to quantity or
amount.” In regard to color space the term uniformity has historically had two
different uses: (1) Absence of variation in terms of a single concept: perceived
color difference between two grades in any direction in space. (2) Absence of
variation in terms of attributes that are perceptually significant but do not
result in perceptually uniform differences, such as blackness or relative content
of unique hues. In the former case the space is uniform in terms of the mag-
nitude of perceived differences but not uniform in terms of blackness or
 UNIFORM AND REGULAR COLOR SPACES 9

redness. In the latter case the space is uniform only in terms of the chosen
attributes but not in terms of perceived differences. It is useful to reserve
the term “uniform” for the former situation and use another term, perhaps
“regular color space” (Hering space for the Hering-inspired version), for the
latter.
As will be seen in Chapter 2, the concept of uniform color solid has a long
history. In the seventeenth century Glisson attempted to develop a gray (light-
ness) scale and three tonal color scales with visually equidistant steps with
which to specify the colors of objects. T. Mayer, J. H. Lambert, and P. O. Runge
in the eighteenth and nineteenth centuries were already thinking in terms of
visually uniform steps between the scale points of their color solids. Mayer
appears to have been the first to propose a three-dimensional color solid. H.
von Helmholtz was the first to attempt to find the relationship between phys-
ical measurements of the stimulus and a perceptually uniform space.
W. Ostwald, apparently through a misunderstanding of Helmholtz’s concept
of brightness, decided to use Hering’s blackness and whiteness as the two
attributes that together with chromatic color, form the color perception. He
used Hering’s equilateral triangular template to arrange all color perceptions
of a given hue. In this template, in the tradition of Runge, W. Wundt, and
Hering, lightness is not an attribute and chromaticness of all full colors (pure
pigment or maximal color) is considered perceptually equal. In regard to chro-
maticness the result is that the perceptual magnitude of chromaticness steps
depends on hue. In addition Ostwald decided that the Weber-Fechner law was
applicable regardless of size of color difference, and he scaled the grades in
the hue triangle accordingly.
Munsell introduced a radical philosophical departure from the German
school by using the three attributes lightness, hue, and chroma. His chromat-
icness measure, the chroma unit, is in principle of equal perceptual magnitude
regardless of hue, and it is defined in terms of (imperfectly defined) constant
perceived lightness. While he originally constrained his color solid into the
form of a sphere, Munsell soon learned from experiments that when building
his system from the central gray midpoint perceptual uniformity was not com-
patible with the complete color solid having a spherical form. The result was
the irregular shape of the Munsell “color tree.” Munsell’s successors continued
to refine the scaling of the three attributes, the last accepted revision being the
Munsell Renotations. In the Renotations perceptual data were “smoothed” to
some degree in terms of psychophysical data.
A major reason for the development of early forms of color solids, as
Chapter 2 will show, was to have a basis for discovering systematic rules of
color harmony. This desire was behind the efforts of Runge, O. N. Rood,
Munsell, Ostwald, and others (Schwarz, 1999). An American version of
Ostwald’s system was called Color Harmony Manual. Even though claims of
having discovered universal rules of color harmony have been discredited,
there has been a continuing discussion on the usefulness of the various systems
10 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

for the purposes of art and design. Ostwald strove to make his system deriv-
able from additive color mixture data, thus developing a system that attempted
to combine perceptual psychology, psychophysics, and harmony.
The Swedish Natural Color System (NCS) is a modern interpretation of
Hering’s ideas. It was derived on a purely psychological basis using presumed
innate concepts of Hering type Vollfarben (full colors) with unique hues, black-
ness, and whiteness. Psychophysical measures were used only to specify color
grade samples exemplifying the system under a specific set of conditions. The
attributes of this Hering or Ostwald type of system are hue, expressed by quad-
rant in terms of one or two unique hues, blackness, and whiteness (or hue,
blackness, and chromaticness). The double-cone geometrical form of systems
such as Ostwald’s and NCS’s appears to imply conventional definitions of the
geometrical dimensions. But by placing all full colors on the periphery of
the central disk of the double cone and a perceptually uniform gray scale on
the central vertical axis, the meaning of the vertical dimension in these systems
is not defined. As a result the steps are not uniform in the sense defined above
but regular. The practical value of such systems must be found in principles
other than uniformity of difference.
The Munsell system on the other hand, as mentioned, is based on the psy-
chological attributes hue, as expressed in terms of five primary hues, value
(lightness), and chroma (saturation). Munsell’s original intent was to repre-
sent a uniform color space. However, by concentrating on planes of constant
hue, he neglected the changes in hue difference as a function of chroma and
lightness between adjacent constant hue planes. A uniform version of the
Munsell system is impossible to fit into a euclidean system as will be shown.
By disregarding the issue of relative perceptual size of hue and chroma dif-
ference steps, the Munsell system is simply accepted as fitting a polar system.
In this system the polar angle, radial distance, and distance from the origin in
the third dimension have defined meanings: hue, chroma, and lightness; but
the units are of different perceptual size (in the case of hue also as a function
of chroma). This was experimentally determined in the 1930s. According to D.
Nickerson’s index of fading formula (1936), one unit of value difference is
equal to two units of chroma difference and, at chroma 5, to three 100-step
units of hue difference.
In 1943, based on the then newly available calculations by D. L. MacAdam
of the object color limits, Nickerson and S. Newhall calculated two three-
dimensional models of the psychological Munsells solid (Fig. 1-2).8 They are
approximations of a uniform psychological solid under two different observa-
tional conditions without among other things, considering the matter of the
relationship between unit hue and chroma differences. They were described
as fulfilling the following requirements: “Dimensional scales . . . calibrated in
perceptually uniform steps; units of the several scales . . . equated; the surface
of the solid . . . represents all colors of maximum saturation; the difference and
volume . . . representative of all colors which are perceptibly different; condi-
tions of stimulation or viewing . . . described; and, finally, the scales . . . stan-
 UNIFORM AND REGULAR COLOR SPACES 11

Fig. 1-2 Models of Nickerson and Newhall’s psychological color solid, based on the Munsell
system. The two figures represent color solids based on large and small perceived differences.
Left: At the level of Munsell Book of Color differences. Right: At the just noticeable difference
level.

dardized in terms of a generally recognized psychophysical system.” B.

Bellamy and Newhall (1942) investigated differences at the threshold level
and reported them in terms of the Munsell attributes. Their results, surpris-
ingly, indicated one unit of value to be equal to eight units of chroma and, at
chroma 6, to 22 units of hue difference, thus indicating a vast change in rela-
tive importance as the differences became small. The shorter version of the
model represents space proportions when viewing differences of the magni-
tude of chroma or value steps. The taller version has the vertical dimension
increased by a factor 4 to indicate the scales when judging samples differing
at the just noticeable difference level. Figure 1-3 illustrates cross sections at
the five basic Munsell hues through the solid, with the inner contours repre-
senting areas covered by actual Munsell color samples. These figures,
seemingly, are the first attempt at realistic (but euclidean) geometrical repre-
sentation of the universe of (relativized) human object color perceptions in
approximations of perceptually uniform spaces.
D. B. Judd and I. H. Godlove, investigating the relationship between hue
and chroma differences used by Nickerson in her formula, discovered that it
is not possible to map the results onto a flat plane. According to this formula
the radial distance covered by the hue differences at any given chroma level
is approximately twice that of a circle, meaning the total hue angle is approx-
12 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

Fig. 1-3 Vertical sections in five different hue planes through the color solid that represents
the Munsell system extrapolated to the optimal color limits. The inner borders delineate the
space filled by samples of the system.

imately 720 degrees. Judd attributed this result to experimental error, and
Godlove (1951) wrote a formula that reduces the magnitude of hue differ-
ences so that the Munsell equal lightness psychological data map onto a plane.
Toward the end of his life Judd reconsidered his view, and there is now sig-
nificant additional evidence indicating that the Nickerson formula is approx-
imately correct. Full clarification requires further psychological scaling.
 UNIFORM AND REGULAR COLOR SPACES 13

As will be shown in Chapter 5, three sets of extensive chroma spacing data,

determined at different times, are not in good agreement, and the implicit
chroma scale of the Optical Society of America Unform Color Scales (OSA-
UCS) system does not agree with any of the three. The psychophysical chroma
scale implicit in the widely used CIELAB color space formula is not in good
agreement with any of the above.9 It is fair to say that we do not have a
reliable chroma scale. Similarly there are no data of hue scaling around a
hue circle at constant chroma and lightness that can be considered reliable
and replicated. Different formulas have been proposed for the weighting of
CIELAB hue differences in recent years and formulas optimal for one set of
data usually perform significantly less well for another. We also do not have a
psychophysical model with known scientific validity for uniform hue spacing.
Similary, elementaristic lightness scales are found to depend on surround
conditions and there is poor agreement between perceived lightness of chro-
matic color patches and measured luminous reflectance.
In 1969 Judd wrote an essay on the subject of ideal color space. His initial
definition of an ideal color space was: “Ideal color space is a tridimensional
array of points, each representing a color, so located that the length of the
straight line between any two points is proportional to the perceived size of
the difference between the colors represented by the points.” A number of
experimental facts, however, led him to conclude that ideal color space by this
definition is impossible. He listed these facts as:

1. Evidence for curvature of color space from the MacAdam ellipse data.
2. Superimportance of hue as indicated in the Nickerson formula.
3. Diminishing returns in color difference perception.
4. Influence of surround color.

Accounting for these problems, he offered the following redefinition of the

concept of ideal color space: “Ideal color space redefined is a tridimensional
array of points, each representing a color, such that all pairs of points sepa-
rated by any fixed distance correspond to pairs of color perceived to differ by
the same amount provided that the appraisal of the perceived size be carried
out with optimal surround colors chosen in accord with Schönfelder’s law10
that the surround be the average of the two colors being compared.” It is
evident that uniformity of difference is the central principle behind both
definitions.
In the 1970s R. M. Evans and B. Swenholt extensively investigated psycho-
logical color space and concluded that to accommodate achromatic surrounds
of varying brightness or lightness, five attributes require consideration and
that, therefore, a euclidean map of color space is a simplification applying to
one surround only.
In its work the Uniform Color Scales Committee of the Optical Society of
America was well aware of the problem of hue superimportance. As men-
tioned, it abandoned separate scaling of color attributes in favor of scaling
14 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

complex chromatic and chromatic/lightness differences. Based on a euclidean

formula fitted to the visual data it proceeded to tile, according to a proposal
by I. Balinkin and C. E. Foss, the implicit space uniformly in twelve directions
from a central midgray. The resulting space combines an irregular shape with
a crystalline interior structure (see Chapters 2 and 7). As a result hue super-
importance was purposely neglected, and the space is not visually uniform in
all directions.
In 1981 G. Wyszecki defined uniform color space as follows: “A uniform
color space is a geometrical representation of color perceptions in a three-
dimensional space in which the distance between any two points can be taken
as a measure of the magnitude of the difference between the two color per-
ceptions that are represented by the two points.” It appears that the concerns
of Judd and Evans about surround and other issues had been shelved.
Today it is evident that there is no single uniform color space and no simple
geometrical model of perceptually uniform space that is more than an approx-
imation. In addition different surrounds and different sizes of intervals on
which the space may be based result in different geometrical forms of the
space and different selections of color chips within the space to represent visu-
ally uniform steps.
A goal of sensory psychophysics is to determine the relationship between
physical stimuli and psychological response. In the case of colors seen as those
of objects, this requires discovering the relationship between the spectral
return of light, reflected from objects, and the psychological response of the
observer. Given the variability in response of observers, this is usually done
for an average color normal observer. Quantitative description of human color
vision used in models consists either of functions representing the average sen-
sitivity of the three cones or color-matching functions that predict if two dif-
ferent spectral power distributions are seen as matching by the standard
observer for whom the color-matching functions apply. The two sets of func-
tions are considered linearly related. Many color space formulas contain addi-
tional suppositions concerning the color vision apparatus, in particular, an
opponent color theory. In the last fifty years the CIE has proposed several
color space and color difference formulas (see chapter 6). Other formulas have
been proposed by other organizations and by individuals. The best of these
formulas explain 65% to 80% of the average variation in perceived differences
of the visual data on which they are based. Significant further improvement is
unlikely without reliable visual hue, chroma and lightness scales and under-
standing of how size of difference affects the implicit color space. Most for-
mulas for object colors developed in the last twenty years are based on the
CIELAB formula, recommended by the CIE in 1976 as a compromise formula
for unification of practice.As will be demonstrated, this formula is quite clearly
not a good basis for color difference calculation. Improvements in the fit of
the formula to visual data since then have been based on statistically deter-
mined mathematical fixes. This text presents new, more detailed understand-
 COLOR SPACE, SENSATION, PERCEPTION, AND AWARENESS 15

ing of the relationship between stimuli conventionally taken as color stimuli

and color perceptions associated with objects, with the potential to provide a
basis for improved formulas.

1.4 COLOR SPACE, SENSATION, PERCEPTION, AND AWARENESS

The terms sensation and perception11 have traditionally referred to immediate

and direct qualitative experiences such as “hard,” “cold,” and “green” in the
former case and complete psychological processes involving implied meaning,
past experience, memory, and strategy in the latter. This view of the visual
system descended from ideas by Descartes who described “three grades of
sensory response. The first is limited to the immediate stimulation of the bodily
organs by external objects. . . . The second grade comprises all the immediate
effects produced in the mind . . . such effects include pain, pleasure, hunger,
colors, sound, taste . . . The third grade includes all the judgments about things
outside us . . .” In recent years it has become quite clear that such a distinction
has little connection with reality. In order to form a perception, we now under-
stand that it is necessary to pay attention to a stimulus and thereby become
aware of it. Experiments have shown that the visual system continuously senses
a large number of stimuli arriving at the retina without the owner of the system
becoming aware of them or having recollection of them. It seems useful to use
the term perception for what results from a given local stimulus after it has
received attention and it has undergone the complete processing resulting in
awareness. A step taken in connection with color scaling is to form judgments
based on perception.
Color perception and the concept of color space and solid are important
components of the not well-defined concepts of awareness and consciousness.
Consciousness remains a mystery but is now being investigated intensively by
neuroscientists, physicists, psychologists, and philosophers. There is a growing
corpus of neurological information regarding the functioning of the visual
system. At the same time we have more than 200 years of investigations of
psychological color space. But there continues to be a black box into which
biologically produced correlates of physical stimuli disappear and out of which
color experiences appear. This situation prevents the development of a con-
vincing model of human color vision and has resulted in the use of growing
numbers of mathematical variables to fit cone sensitivity or color matching
function data to perceptual data. It is generally accepted that all information
derived from radiant energy required for us to experience form, color, and
motion passes through the filter of the three cones. If this is an ultimate truth,
then we must look to neurophysiology of the retina and pathways in the brain
to provide more information on the processes as a basis for better models.
What we have seems promising and, at the same time, is unsatisfactory. This
16 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

assessment is based on the unproved assumption that color experiences are

directly derivable from the neurophysiological functioning of certain cells in
our visual system, an idea that has, more or less directly, informed efforts
toward a uniform psychophysical color space for the last century. However,
there is a significant group of scientists and philosophers disagreeing with it,
and the issue must be considered open.
In the 1940s color space and difference research in the United States prof-
ited from the historical curiosity of a search for work not directly connected
with war effort (Nickerson, 1977). Early results generated their own momen-
tum, and work continued in the 1950s and 1960s through the efforts of a few
dedicated individuals in a committee of the Optical Society of America. In the
1960s, with growing capabilities for industrial reflectance measurement and
color calculation, color technologists in colorant producing and using indus-
tries around the world became increasingly interested in the possibilities of
objective color quality control and provided impetus for new work. These
efforts have resulted in the level of success mentioned above, based on reli-
able reflectance measurement techniques and the developments in color
science to be discussed below. Since the mid-1980s the major activity has
shifted to academic institutions. More recently lack of funding and new and
seemingly more exciting fields of research in color have slowed the pace of
color space and difference research work appreciably. It seems that we must
rely again on a few dedicated individuals, interested in pushing the frontier in
this field for the sake of advancing toward the distant goal of understanding
qualitatively and quantitatively, in very limited situations, the relationship
between visual stimuli and the resulting color and color difference perceptions.

1.5 PLAN OF THE BOOK

Chapter 2 begins with an attempt at a general definition of the meaning of

color space and color solid. It is followed by a historical survey of ideas about
color order beginning with ancient Greek philosophers. Given the paucity of
surviving documents, our knowledge in this area is likely incomplete. The
survey continues through the Middle Ages and the Renaissance into the Age
of Enlightenment. Three-dimensional color solids began making an appear-
ance in the eighteenth century. The contributions of psychophysics, starting in
the midnineteenth century, to the matter at hand are discussed as are devel-
opment in understanding of human color vision and of the colorimetric
system. Brief discussions of the systems of Hering, Munsell, Ostwald, the
German DIN 6164, the Optical Society of America Uniform Color Scales, the
Swedish Natural Color System, and others, including systems used in video
display, are presented to bring the reader to the present in this multimillennia
pursuit of ordering our color perceptions.
Chapter 3 offers a survey of psychophysics as relevant to the color-
ordering enterprise. Many of the problems and complexities of psychophysics
 PLAN OF THE BOOK 17

are touched on including theories of categorization, relationship of differences

and magnitudes, uni- and multidimensional scaling methods, and the relation-
ship between psychological and psychophysical color spaces.
The theme of Chapter 4 is perceptual color attributes and how they are
scaled. It concentrates on perceptual scaling only. Views regarding color attrib-
utes have a history of their own. On the one hand is the physics inspired set
of hue, saturation, and brightness or lightness introduced by Newton and
Helmholtz, on the other, Herings “natural” system of hue, whiteness, and
blackness. A large portion of the chapter is given to data of perceptual scaling
of color attributes, including location of unique hues and distances between
them. The paucity of extensive sets of global scaling data and the lack (for
unknown reasons) of close agreement among those few that exist is com-
mented on.
In Chapter 5 the perceptual scales of Chapter 4 are related to physical def-
initions of color stimuli such as reflectance or spectral power distribution data.
This requires brief discussions of photometry and colorimetry as well as psy-
chophysical spaces such as cone response, tristimulus, and opponent color
spaces. The relationship between color matching and color appearance is
touched on, as is placement of unique hues in psychophysical spaces and cur-
vature of lines connecting blues of constant hue in CIE-based opponent color
diagrams. The chapter closes with a discussion of the number of colors we can
distinguish.
Chapter 6 contains all major historical steps in the effort of finding psy-
chophysical formulas attempting to describe uniform color space, beginning
with Helmholtz’ line element and ending in the present. The chapter ends with
a brief comparison of color and spectral spaces as well as a comparison of per-
formance of various formulas against the Munsell system and the RIT-DuPont
data exemplifying global color and small color difference data.
Chapter 7 contains more extensive descriptions and comparative analysis
of three major color order systems: the Munsell and OSA-UCS system and
the NCS system. The former two are attempts at a uniform color space while
the latter is an implementation of Herings “natural color system.” It is demon-
strated that neither constant value planes of the Munsell system nor constant
lightness planes of the (experimental results of the) OSA-UCS can be iso-
morphically plotted on a euclidean plane.
In Chapter 8 color differences have been scaled at many different levels,
from color-matching error and threshold differences to small and large
suprathreshold differences. In this chapter important data from each category
are compared in the cone sensitivity and the tristimulus spaces for agreement
and discrepancies. One of the comparisons involves the magnitude of Weber
fractions, another the direction of the unit chromatic contours in the spaces.
Many issues are found to be unresolved because of lack of reliable, replicated
data.
Chapter 9 draws conclusions from the facts of the previous chapters by
attempting to answer 13 questions related to color scaling and uniform and
18 THE CONCEPT OF COLOR SPACE AND COLOR SOLID

regular color spaces. An approximation of a uniform color space at the level

of small color differences is shown as a counterpoint to the figure of the
Nickerson-Newhall uniform color space in this chapter (Fig. 1-2). A research
program is proposed in some detail to establish reliable, replicated data that
can be used to determine the properties of uniform color spaces for a given
viewing and surround situation at different levels of size of difference.
 Chapter 2
Historical Development
of Color Order Systems

2.1 COLOR AND COLOR ORDER SYSTEMS

One of the definitions in the Oxford English Dictionary, not without its diffi-
culties, of the word color is “The quality or attribute in virtue of which objects
present different appearances to the eye, when considered with regard only to
the kind of light reflected from their surfaces.”The definition is symptomatic of
the age-old problems of clearly describing this psychological quality. But we all
(those of us with normal color vision) know colors when we encounter them.
In 1912 the philosopher B. Russell expressed himself on the subject as follows
“. . . truths about the colour do not make me know the colour itself better than
I did before . . . I know the color perfectly and completely when I see it. . . .”
The conventional view of colors is that they are the result of standard
stimuli associated with them. I. Newton’s seven primary hues (R O Y G B I
V) are the result of spectral stimuli of particular wavelengths (as seen in a
dark surround). In color technology certain pigment or dye combinations
result in certain color perceptions when viewed under a standard light in a
light booth. That the appearance may change significantly if we change the
light source or the surround is described in terms of color inconstancy. The
idea that there is a standard stimulus for a given color perception has often
been questioned (independent of the issue of metamerism). In the 1960s the
inventor of the Polaroid photographic process, E. Land, sharply delineated the
limitations of such views. In natural scenes the colors assigned to a given spec-
tral power distribution entering the eye depend on the spectral structure of
the total visual field according to rules that are not yet understood. It does not

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

19
20 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

seem appropriate to talk about color illusions when the implied stimulus/per-
ception association is violated. Colors perceived in any given situation are the
real colors. Our mind constructs them on the basis of its interpretation of the
total visual field as well as of items in this field that appear to belong together.
This view causes severe problems for the idea of a color order system in which
the specimens are defined by their reflectances. On the other hand, there
appears to be a finite limit in number of color perceptions that the color
normal observer can experience when seen as colors of objects.There is a legit-
imate question as to how one can systematically demonstrate these percep-
tions. The answer is that we have to select a standard set of conditions in which
to view the objects in order to be able to more closely relate stimuli and per-
ceptions. This is the approach taken implicitly or explicitly by all developers
of physical color order systems. After the materials and the observation situ-
ation have been fixed the variability of color normal observers comes into
focus. As a result more recent efforts in development of color order systems
always involved average data from many observers. From this perspective,
extensive relativization of the viewing circumstances to tightly controlled con-
ditions, relating color perceptions to particular spectral stimuli has a degree of
validity. Any claim to general validity is erroneous, however.
There are two situations in which it is meaningful to associate a color per-
ception with a specific stimulus:

1. A specific set of conditions fixes the association, but it is understood that

the ensuing color experience can be the result of many other sets of
stimuli and surround conditions. All the associations are taken to repre-
sent the universe of possible object color perceptions.
2. The surround conditions represent a reasonable and practical limitation
for the purposes of controlling the industrial production of materials
with reflectance/transmittance properties making the objects seen as
colored.

The latter situation deals with two issues:

1. The initial colorant formulation for objects needs to meet criteria of

closeness to the reference sample and, perhaps, degree of color con-
stancy (lack of perceived change of color of formulation under different
light sources).
2. The degree of perceived difference between the approved formulation
sample and subsequent repeated manufacture of the material should be
minimal.

In both cases it appears justified to use simplified surround conditions for

visual tests. Objective test methods must reflect these conditions.
Color order systems that, as used here, include color appearance systems
 FROM ANCIENT GREECE TO THE MIDDLE AGES 21

are logical arrangements of color chips and/or geometrical and numerical

arrangements of symbols for such chips. One system (NCS) is based on mental
references defining the color appearance.The chips themselves are symbols for
color perceptions that can be created when the average color normal person
views them under specified conditions of lighting and surround. Color appear-
ance systems, specifically, are atlases of collections of color chips or prints.1
The remainder of this chapter presents the historical development of color
order systems in the Western world.

2.2 FROM ANCIENT GREECE TO THE MIDDLE AGES

Pre-Platonists
Among the oldest extant uses of color words are those by the poet, theolo-
gian, and natural philosopher Xenophanes, active in the sixth century bc. He
explained all things to have come from water and earth and commented on
the rainbow: “And she whom they call Iris, she too is actually a cloud, purple
and flame-red and yellow to behold.”2
Pythagoras (ca. 582–507 bc) philosopher and founder of a religious broth-
erhood is credited by his followers as having discovered that the relationship
between musical notes could be expressed with numbers, as could any other
relationship. Of considerable importance to the Pythagoreans was the tetrac-
tys, their name for the sum of the first four numbers, regarded as the source
of all things. According to Philolaus, a disciple of Pythagoras, he equated colors
with the number 5. Plutarch remarks on the views of the Pythagoreans on
color: “[They] called the surface of a solid chroma, that means color. Addi-
tionally, they named the species of color, white, black, red and yellow. They
looked for the cause of the differences in color in various mixtures of the ele-
ments, the manifold colors of animals, however, in their nutrients as well as
the climatic regions.”3
Empedocles (ca. 492–432 bc), a Sicilian philosopher and politician, is cred-
ited with the view that everything in existence is composed of four indestruc-
tible elements: earth, fire, water, and air. On color Empedocles is quoted as
follows: (after Simplicius, 28) “As when painters decorate temple-offerings
with colors—men who, following their intelligence, are well-skilled in their
craft—these, when they take many-colored pigments in their hands, and have
mixed them in harmony, taking more of some, less of another, create from
them forms like all things. . . .” (after Aetios, 132) “Empedokles explained
color as that fitting into the pores of the visual organ. The multiplicity of colors,
so he said, derive from the fixed mixtures of the elements. That there are four
colors, exactly as many as elements: white [leukhyn], black [melan], red
[erydron], yellow-green [ochron].”4
Democritus (ca. 460–? bc) expanded the atomistic theory of his predeces-
sor Leucippus. Among many other works he is reported to have written a text
22 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

On colors, not extant. Grammarians credited him as the inventor of then

unusual words, among them one with the meaning “change of color.” He is
reported by Galen to be the author of “By convention there is color, by con-
vention sweetness and bitterness, but in reality only atoms and void.” On
primary colors and color mixture Democritus is quoted by Theophrastus (97):
“According to Democritus there are four primary colors. White [leukhyn], he
explained, is the smooth, polished. . . . The black color [melan], on the other
hand, consists of (forms) of an opposite nature: rough, nodulated and uneven.
. . . Red [erydron] he said to consist of similar but larger (forms), as warmth.
. . . The green color [khloron], on the other hand, consists of hardness and
emptiness, as a mixture of both. . . . The primary colors he said to be based on
these forms. Each of them being purer the less it is composed of mixed forms.
The other colors he said to derive from mixtures of the (four primary colors).”5

Plato
Plato (427?–347 bc), philosopher and mathematician, friend and pupil of
Socrates, founder of the Academy, was one of the preeminent figures in Greek
philosophy. His influence remains active today. Plato made several statements
about colors in his dialogues. Concerning primary colors, and their mixtures
his key statement is found in Timaeus (67, 68):

. . . we ought to term white that which dilates the visual ray, and the opposite of
this black. . . . in (the eye) the fire, mingling with the ray of the moisture, pro-
duces a color like blood, to which we give the name red. A bright hue mingled
with red and white gives the color auburn [xandon]. The law of proportion,
however, according to which the several colors are formed, even if a man knew
he would be foolish in telling, for he could not give any necessary reason, nor
indeed any tolerable or probable explanation of them. Again, red when mingled
with black and white, becomes purple, but it becomes umber [orphninon] when
the colors are burned as well as mingled and the black more thoroughly mixed
with them. Flame color [pyrron] is produced by a union of auburn and dun
[phaion], dun by an admixture of black and white; and pale yellow [ochron] by
an admixture of white and auburn. White and bright meeting and falling upon a
full black, become dark blue [kyanoyn], and when the dark blue mingles with
white a light blue [glaykon] color is formed as flame color with black makes leek-
green [prasion]. There will be no difficulty in seeing how and by what mixtures
the colors derived from these are made according to the rules of probability.

Plato’s color mixture scheme is shown graphically in Fig. 2-1. Note that the
classification into primary, and so on, colors is by the author.

Aristotle
Aristotle (384–322 bc) was a student of Plato at the Academy. After teaching
Alexander the Great he founded a school in the Lyceum in Athens. Together
with Plato, Aristotle had an enormous influence on philosophical thinking
 FROM ANCIENT GREECE TO THE MIDDLE AGES 23

Fig. 2-1 Plato’s color mixture scheme. The four primary experiences are bright, white, red and
black. Additions of these in various combinations form the secondary and later stage mixture
colors.

during the next 2000 years. In regard to basic colors and color scales
Aristotle expressed himself on more than one occasion and not always in the
same way.
In Sense and Sensibilia (442a20 ff): “Savors and colors contain respectively
about the same number of species. For there are seven species of each, if, as
is reasonable, we regard gray as a variety of black [melanon] (for the alterna-
tive is that yellow [xandon] should be classed with white [leukhon], as rich with
sweet); while crimson [phoinikoyn], violet [aloyrgon], leek-green [prasinon]
and deep blue [kyanoyn], come between white and black, and from these all
others are derived by mixture.”
Earlier he stated: (439b20 ff):

Publisher's Note:
Permission to reproduce this quote
was not granted by the copyright
holder. Readers are kindly requested
to refer to the printed version
of this article.
24 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Publisher's Note:
Permission to reproduce this quote
was not granted by the copyright
holder. Readers are kindly requested
to refer to the printed version
of this article.

On Colors is a text attributed to Aristotle, but by some writers also to

Theophrastus, Aristotle’s successor at the Lyceum, and believed to represent
Aristotle’s final thinking on the subject. Here he stated (1–3):

Publisher's Note:
Permission to reproduce this quote
was not granted by the copyright
holder. Readers are kindly requested
to refer to the printed version
of this article.

Ptolemy
This Greco-Egyptian mathematician and astronomer (second century ad) is
reported to have mentioned in the lost first book of his Optics a list of twelve
 FROM ANCIENT GREECE TO THE MIDDLE AGES 25

simply seen colors. Eleven of these are listed in a biography of Ptolemy found
in the library of the ninth-century Byzantine scholar and patriarch of Con-
stantinople Photius.6 They are, in no particular order: black, white, orange
[xandos], dun, gray [phaion], yellow [ochros], red [erythros], blue [kyanos],
purple [halurgos], shining or bright [lampron], and dark brown [orphinon].

Ancient Greek Thinking on Color Species and Color Order

The idea that there are four material elements and correspondingly four fun-
damental colors belongs to Empedocles. It proved to be influential for many
centuries. The quote about painters attributed to him indicates that naturalis-
tic painting and the mixture of pigments to achieve certain colors were well
established in Greece at his time. Democritus who explained the world in
atomistic terms is reported by Aristotle to have said that color does not have
existence because it is generated by the turning of atoms. Democritus changed
the name of the fourth, yellow/green elemental color from Empedocles’s
ochron to khloron.
Several other pre-Socratic philosophers are reported to have written texts
on natural philosophy and are likely to have discussed senses. But their works
are lost. Among them are Heraclitus, Alcmaeon, Anaxagoras, Diogenes, and
Leucippus.
From Plato’s statements it is evident that his elemental colors bright, white,
red and black are not colorants but ideas of color fundamentals. From these
in secondary, tertiary, and quaternary mixtures the next eight colors are
created (see Fig. 2-1). Additional colors are generated from further mixture.
There is clearly an awareness of lightness from the sequences “auburn plus
white makes light yellow” and “dark blue plus white makes light blue”, as well
as “purple burned makes umber”. But there is no indication of a systematic
arrangement by lightness.
It is apparent that Aristotle was aware of a fundamental difference between
mixing lights and mixing pigments. In Sense and Sensibilia he offers seven fun-
damental colors seemingly ordered by brightness. In On Colors they appear to
be reduced to two, white of air, water and earth, and yellow of fire. Black is the
result of transmutation. Here he also showed his awareness of differences in
intensity and brightness that can be present in colors of a given hue. Influenced
by Pythagoras and his followers, he believed that colors appearing harmonious
together are formed by simple ratios. Like Plato he was overwhelmed by the
multitude of colors and could not see a way of organizing them systematically.
Ptolemy’s list appears derived from Plato’s except that orange [xandos] is
added, as is gray [phaion]. Missing is leek green [prasinon] and flame color
[pyrrhon], and there is only one bluish color. It is a simple list without any
apparent order.
Pliny (ad 23–79), in a history of Greek art indicates that the painter
Polygnotus, active in the fifth-century bc, added light ochre and other pigments
to the standard early colors of painting, black, white, and red. He is reported
26 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

to have painted women in light robes and their jewelry in various bright colors.
However, the idea of light and shadow was not yet understood. Pliny puts the
invention of cinnabar pigment into the same time period. Light and shade
appear to have become common in Greek painting one or two generations
later. Invention of further new pigments in the fifth to second centuries bc
made more and more naturalistic painting possible and required more exten-
sive practical knowledge in pigment mixture.
Of considerable and ongoing controversy is Pliny’s statement that the most
refined classical Greek painters only used four colors in their work: white,
black, yellow, and red.7 Blue pigments must have been known to the Greeks
through trade with Egypt and other nations in the region. Traces of blue pig-
ments have been discovered on statuary and on temple friezes. Works of the
celebrated classical painters, like Apelles, have not survived. What is reputed
to be a copy in mosaic of a painting by Philoxenos (4th c. bc) was found in the
ruins of the House of the Faun (estimated to have been made ca. 100 bc) in
Pompeii. It represents Alexander the Great confronting Darius at the battle
of Issus. The colors of its tesserae are muted, beiges, browns, and grays, without
any blues or greens, thus being in agreement with a four-color painting theory.
The sacral colors of the ancient Greeks are white, the color of festivities,
black, the color of mourning, and various reddish shades from scarlet to violet.8
White bulls or lambs were often used in sacrifices. Red, as the color of blood,
invoked both death and life.
E. Veckenstedt has shown that the classical Greek epic writers used some
140 color words, while the philosophers of the same time period only used
approximately 50 words. He also showed that of the words used by the epic
writers and the philosophers nearly twice as many tended toward white than
toward black (Veckenstedt, 1888). Clearly, in the practical world many color
experiences were distinguished by name even though philosophers mentioned
only relatively few. At the same time there was considerable confusion as to
the exact meaning of given color names, well understandable in the absence
of color standards. B. Berlin and P. Kay (1969) have classified Homeric Greek
as representing stage IIIb in their scheme of development of basic color terms,
namely having the basic terms of black, white yellow, red, and blue. By the
time of the later classical epic writers the color palette had become quite
extensive, with many terms indicating mixed hues and toned colors. There is
no record, however, of a comprehensive effort to sort and order the multitude
of colors.
A perennial problem is that of interpreting the meaning of color names.
Without having specific examples for the meaning of various names, later
interpretations vary considerably. It is evident that throughout the classical
history of Greece what developed into color names had originally more
general meanings. The situation became more complicated in Roman times as
different translators of Greek texts sometimes used different Latin words for
a given Greek color word. A particularly confusing example is the word
glaukos, variously translated as light blue, grayish green, or gray, in one case
 FROM ANCIENT GREECE TO THE MIDDLE AGES 27

as flashy. As seen later, in the late Middle Ages it briefly assumed also the
meaning of yellow.

Color Order in the Middle Ages

According to the record Romans made no fundamental contributions to the
subject of color order. Lucretius when discussing vision in De rerum natura
(The way things are), his poetic account of Epicurean philosophy, essentially
followed the atomic theory of Leucippus and Democritus as commented upon
by Epicurus.
In the forth-century ad the translator and commentator Chalcidius (life
dates not known) produced the translation of Plato’s Timaeus dialogue most
widely known in the Middle Ages. His text includes commentary. In the com-
mentary on section 67ff of the Timaeus, Chalcidius described a five-term scale
of basic colors as follows: white–yellow–red–blue–black, apparently the first
written record of a reordering of the classical Greek fundamental colors into
a single tonal scale with red in the middle.

Avicenna and Averroës

The step to multiple tonal scales is found in documentary evidence related to
the Persian philosopher and physician Avicenna (Abu Ali al-Husayn ibn Sina,
980–1037). Avicenna was knowledgeable in Aristotle’s writings. In his specu-
lations on the human soul, influenced by Aristotle, he also addressed issues of
color vision. In his manuscript, translated into Latin in the twelfth century,
Liber de anima seu sextus de animalibus (Book of the soul) he may have been
the first to describe tonal scales of (more or less) individual hues. Changes
along Avicenna’s tonal scales9 involve, in modern terminology, most likely both
lightness and chroma. He described three such sequences:

1. From subpallidum to pallidum (a gray or perhaps yellow scale, the gen-

erally accepted translation of pallidus is pale, wan, and its meaning as a
hue term is in dispute)
2. From subrubreum to rubreum (a red scale)
3. From green to indigo (a scale involving hue as well as lightness and
chroma changes)

The prefix “sub-” has the meaning of below, and the first two scales refer to
perhaps grays or browns to yellow, respectively wine red to full red. The third
scale travels across an expanse of hue as well as lightness, from a medium
green to a dark reddish blue. These appear to be the first instance of a color
arrangement that requires at least two dimensions to represent it.
The Spanish-Arabic philosopher and commentator ibn-Rushd
(1126?–1198?), known under the Latin name of Averroës, wrote extensive
28 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

commentaries on Aristotle, among many other texts. In his commentary on

Sense and Sensibility he introduced terms that were translated into Latin as
remittere, with a meaning to yield, to abate, and intendere, with a meaning to
spread, direct. They were applied to describe an interpretation of Aristotle’s
thesis on colors. Accordingly, as Bacon later expressed it: “. . . as Aristotle
means, the causes generating colors are elementary qualities that are aug-
mented and abated in their generations through the power of brightness”
(Parkhurst 1990). Applied to chromatic colors, it may mean tonal scales, and
to black and white, a gray scale.

Eraclius
Eraclius is believed to have been an Italian monk. Nothing is known about his
life dates or his activities. There are several manuscripts in Latin in existence
that are copies of two books in rhymes and one in prose, less certainly by him,
describing the manufacture of colorants for artistic purposes. Stylistic and
content comparisons have resulted in estimates that he lived in the tenth
century. In section 50 of the prose manuscript Eraclius10 discusses the “various
kinds and names of the principle and intermediate colors . . .”:

Of colours, some are white and some are black. . . . The intermediate colours are
red, green, yellow, purple, prasinus [leek green], azure and indicus [indigo], which
are each of them, in themselves, beautiful; but are more so when mixed, because,
by their variety, they give beauty to one another. And then, in composition, they
have a different hue, . . . colours of different kinds are mixed together, in order
that they may partake of the nature of the others as well as their own, and make
as many, and beautiful, and pleasing, varieties as possible. In this mixture, and in
the order in which one is laid over another in painting, great skill is excercised.

In section 58 he describes how shading and highlights are best achieved for
various colors: “And note that you must shade azure with black; and lay on
the lights with white lead. . . . If you wish to make a colour like lily green, mix
azure with white lead; shade it with azure; lay on the lights with white lead;
and when it is dry, cover it over with clear saffron.”
It is evident from this that Eraclius was well acquainted with color mixing
and painting technology. Mixing of pigments, lasing, and toning in black and
white directions were standard procedures. However, his color sequence does
not indicate a concern with systematic color arrangement.

Theophilus
Explicit tonal scales going from the full color of a bright pigment in both direc-
tions toward white and black were first described by the German Benedictine
monk Theophilus (1080?–1125?), also known under the name Rugerus. In
approximately 1122 he wrote a treatise, De diversis artibus (The various arts)
in which he describes for the benefit of brother monks at some length techni-
 FROM ANCIENT GREECE TO THE MIDDLE AGES 29

cal details for painting, glass making, and metalworking as practiced in his
time. It seems likely that Theophilus was acquainted with the works of Era-
clius. In Book I of his manuscript he describes how to mix colorants for paint-
ing flesh tones with their shadows and highlights as well as facial and head
hair. He also describes how to mix colorants for the painting of draperies in
wall and ceiling painting and how to imitate the rainbow in painting. For flesh
color he begins with heated lead white, having turned yellowish, to which
unheated lead white and vermilion (mercuric sulfide, red pigment) is added
until flesh color is attained. More white is added for light faces, for pallid ones
green earth instead of vermilion. Theophilus proceeds with first and second
shadow colors, progressively darker, with first and second rose colors and a
dark red, first and second highlight colors as well as a dark gray used for paint-
ing eyes. Thus Theophilus describes scales of colors centered on average flesh
color and going into lighter, darker, redder, and greener directions. For
example, the second shadow color for flesh is described as follows:

Afterwards take the (first) shadow color for flesh which has been referred to
above, and mix with it more green earth and burnt ochre so that it is a darker
shade of the former color. Then fill the middle space between the eyebrows and
eyes, under the middle of the eyes, near the nose, between the mouth and chin,
on the down or beards of young men, on the half-palms toward the thumb, on
the feet above the smaller areas of relief, and on the faces of children and women
from the chin right up to the temples.

Similarly, for painting drapery various tonal scales are described, for example,
for a greenish yellow hue:

Mix pure viridian (green) with yellow ochre so that the yellow ochre predomi-
nates, and fills the drapery. Add to this color a little sap green and a little burnt
ochre and make the drawing. Mix white with the ground-color and paint the first
light areas. Add more white, and paint the lighter areas on top. Mix with the
above shadow-color more sap green and burnt ochre and a little viridian and
make the shadow on the outside. . . . Mix dark blue with white in the above way.
Similarly mix black with white. In the same way mix yellow ochre with white and
for its shadow add a little burnt ochre.

The colors of the rainbow he describes as follows:

The band which looks like a rainbow is composed of various colors: namely ver-
milion and viridian, also vermilion and dark blue, viridian and yellow ochre, and
also vermilion and folium [a vegetable red lake]. . . . Then mix from vermilion
and white whatever tones you please so that the first contains a little vermilion,
the second more, the third still more, the fourth yet more, until you reach pure
vermilion. Then mix with this a little burnt ochre, then burnt ochre mixed with
black and finally black. . . . You can never have more than twelve of these strokes
in each color range. And if you want these many so arrange your combinations
that you place a plain color in the seventh row.
30 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Here Theophilus clearly describes twelve-tone tonal scales for various pure
color pigments that represent the most intense color available for a given hue.
Presumably, the scale colors are to be mixed so that the steps appear approx-
imately even. In this fashion he extends Avicenna’s scales in a systematic
manner, however, without proposing a formal arrangement to place these
in.

Learned discussions concerning color scales continued without adding sub-

stantially to what Avicenna, Eraclius, and Theophilus had provided. In the later
twelfth century the Sicilian physician Urso de Salerno11 commented in a dis-
cussion about the four element-related basic colors that there were many and
often unnamed intermediate colors and rather than name them a good painter
could produce them by mixture from the four elemental colors.
Albertus Magnus (ca. 1200–1280) Dominican monk, philosopher, teacher,
and saint had considerable interest in optics and the sense of vision. He added
to Avicenna’s scales a number of colors he believed were missing: fuscum
(dark colored) to the pallidum scale; croceum (saffron colored, golden), pur-
pureum (purple), and indicum (here considered to be a red color) to the red
scale; and viride clarum (bright green) and viriditas intensa (intense greenness)
to the green scale.12

Grossteste and Bacon

Robert Grosseteste (ca. 1168–1253) was an English Franciscan monk, later to
become chancellor of Oxford University and bishop of Lincoln. His extensive
writings include works on optics and color. By postulating a direct path from
the material world to its essential nature through optics, he legitimized exten-
sive studies of optical phenomena. In De colore (On color, ca. 1233), one of
four books on optics, he described a seven-color scale (colors unnamed) cred-
ited to Aristotle and Averroës. These colors [presumably including black and
white] are related to bright light [lux clara] by remission and, implicitly, to
darkness [lux obscura] by intention (Parkhurst, 1990).
Roger Bacon (1214/20–1292), also a Franciscan, Aristotelian philosopher
and scientist, lived a generation after Grossteste. He wrote three different, if
related, texts with sections on color: Liber de sensu et sensato (Book of the
senses and sensibilities, ca. 1255, attributed to Bacon), Opus majus (part of a
projected encyclopedia), and De multiplicatione specierum (On the multipli-
cation of species).13 Bacon attempted to apply the second-century Greek
philosopher Porphyry’s system of predicables, a logical sequence of attrib-
utes—genus, species, difference, property, and accident—to the problem of
color. He applied the term genus to a Chalcidian list of five fundamental color
properties: whiteness [albedo], yellowness [glaucitas], redness [rubedo], blue-
greeness [viriditas], and blackness [nigredo]. Each of these has a range of color
species attached to them: for example, glaucitas is found in lividus, flavus,
glaucus, ceruleus, pallidus, and citrinus (see below for translation). Bacon dis-
cussed at length Grosseteste’s views on color and their relation to lightness
 FROM ANCIENT GREECE TO THE MIDDLE AGES 31

and darkness. He used the specific Latin term gradus (step, gradation) to
describe color variations representing difference in the Porphyry sequence.
In De sensu et sensato Bacon included a list of twenty colors, sorted in
approximately tonal sequence, he considered important and explained their
meaning as derived from older sources (interpretations of Bacon’s meaning
are in parentheses).

candidus shining white

albus white, dead white
lividus incomplete white (ivory, oatmeal)
flavus like lividus (pale yellow)
glaucus yellow
ceruleus between glaucus and citrinus (color of beeswax)
pallidus between citrinus and ceruleus (pale)
citrinus orange, between glaucus and puniceus
puniceus (reddest of the yellows)
rufus (golden, scarlet)
croceus saffron colored, blood red
rubeus true red, the median color between white and black
rubicundus darker, more bluish than rubeus
purpureus (purple, violet)
viridis (blue-green)
venetius between true blue and black (dark blue)
lividus (dark gray?)
lazulus (dark blue)
fuscus (dark colored, no specific hue)
nigrum black
Bacon was on the threshold of a more complete understanding of color phe-
nomena, without being able to establish a clear system.

Dietrich von Freiberg

The German Dominican monk Freiberg wrote circa 1310 a manuscript De
iride (About the rainbow) in which he provided the first accurate explanation
for the rainbow phenomenon. He argued (against Aristotle) that chromatic
colors could not be produced from mixture of white and black: “. . . a mixture
of white and black does not produce anything except a remission of the white
and black from their perfections” (various grays, as well known by artists;
Wuerschmitt 1914). He also used the term glaucus to denote yellow, as did
Bacon and the English encyclopedist Bartholomew Anglicus in 1250
(Parkhurst, 1990).

Medieval Progress toward a Systematic View of Colors

Chalcidius, the translator and commentator of Plato, introduced a tonally
ordered scale of five fundamental colors that was influential until the late
32 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Renaissance (see d’Aguilon and Kircher below). Avicenna’s attempt at order-

ing colors, influenced by Aristotle, resulted for the first time in scales that
modulate a given color along a more or less tonal path. Albertus Magnus com-
mented on the need to add additional steps to Avicenna’s scales. Roger
Bacon’s sustained efforts at color order did not result in significant progress.
It was the practical Theophilus who, based on needs of painters and artisans
in enameling to create life-like images, proposed the toning of a given pigment
color in twelve systematic steps toward white and black. If he saw in this
approach the possibility of a systematic description of the world of colors, he
did not say so in his writings.

2.3 COLOR ORDER IN THE RENAISSANCE

Despite the flowering of the Renaissance on both sides of the Alps and the
writing of books on painting (Leon Battista Alberti, 1335, and Cennino
Cennini, ca. 1400), there was only modest progress in thinking on the subject
of color scales until the sixteenth century. Cennini mentions a list of seven
colors (but not as a scale) that appear to derive from Aristotle: “Know that
there are seven natural colors, or rather four actually mineral in character,
namely, black, red, yellow and green; there are natural colors but need to be
helped artificially, as lime-white, blue-ultramarine, azzurite, giallorino”
(Cennini, ch. 36).14 Cennini described and recommended a style of painting in
which the full pigment colors represent the darkest colors in a picture and gra-
dations are made exclusively with the addition of white (known as the Cennini
style; Hall, 1992). Alberti offered a painter’s view on colors. While he discussed
(in the Latin version of his manuscript) various views on color of philosophers
and experts, in the Italian version he said: “I speak here as a painter. . . .
Through the mixing of colors infinite other colors are born, but there are only
four true colors—as there are four elements—from which more and more other
kinds of colors may be thus created. Red is the color of fire, blue of the air,
green of water, and of the earth gray and ash [bigio et cenericio]. . . . Therefore,
there are four genera of colors, and these make their species according to
the addition of dark and light, black or white. They are thus almost innu-
merable. Therefore the mixing of white does not change the genus of colors
but forms the species. Black contains a similar force in its mixing to make
almost infinite species of color.” The interpretation of Alberti’s choice of the
earth-related fourth primary chromatic color is difficult. It has been described
as a dull yellowish gray (Gavel, 1979). It appears that the association of his
primary chromatic colors with the four elements was for Alberti more
important than a system that would recognize yellow as primary. However, he
clearly envisaged the systematic toning of the primary colors toward black and
white.
The painter and scientist Leonardo da Vinci (1452–1519) proposed a sort
of tonal scale of six colors: “The simple colors are six, of which the first is white,
 COLOR ORDER IN THE RENAISSANCE 33

although some philosophers do not accept white or black in the number of

colors, because one is the origin of all colors and the other their absence. But
as painters cannot do without them, we include them in the number of the
others, and say that in this order white is the first among the simple, yellow is
the second, green is third, blue is fourth, red is fifth, and black is sixth.”15
Leonardo’s hue sequence clearly considers the hue circle. It is obvious that
Leonardo, like Alberti, was also well versed in tonal expansions achievable
with the addition of white and/or black pigments to the pigments represent-
ing the four simple chromatic colors and their mixtures.
A curious twelve-step scale based on brightness and without regard to a
systematic arrangement of hues was proposed by the Italian neoplatonic
philosopher Marsilio Ficino (1433–1499) in a letter to a friend.16 “In light there
are many ideas of colors as there are colors in objects. At the lowest degree
where it is communicable, there is the idea of black, at the second the idea of
brown, at the third dark yellow, at the fourth dark blue and green, at the fifth
sky blue and sea green, at the sixth full red, at the seventh light red, at the
eighth saffron yellow, at the ninth white, at the tenth the transparent or
the shining, at the eleventh the brilliant, and finally, there is the idea of
splendor.”
In 1528 the Italian philologist and poet Antonio Telesio (1482–1534) pub-
lished in Venice a small book Antonii Thylesii de coloribus libellus (Antonio
Telesio’s little book on colors). It is reprinted in its entirety in the original
Latin in Goethe’s Geschichte der Farbenlehre (1810). Telesio described twelve
basic colors, not in any particular order. Aside from the differently named
black only three colors are identical to colors in Bacon’s list. For each of these
basic colors Telesio offered a historical commentary and several related color
names. The list17 and some related color names are as follows:

category name related color names

1. Coeruleus (blue, sky blue) indicum, cyaneum, venetus, blavus
2. Caesius (bluish gray, gray) glaucus, baios, charopon
3. Ater (black, dead black) niger, anthracinus, fuscus
4. Albus (white, dead white) pallidus, candidus, leucophaeus
5. Pullus (dark colored, grayish black) impluviatus
6. Ferugineus (rust colored, dusky) hyacinthus
7. Rufus (red, ruddy) rutilum, russum, sanguinatus
8. Ruber (red, ruddy) purpureus, xerampelinus
9. Roseus (rose colored) incarnatus
10. Puniceus (purple) spadiceus
11. Fulvus (yellowish brown, tawny) croceus, luteus, flammeus
12. Viridis (green) prasinus

The list is remarkable for its focus on desaturated, natural colors and the
absence, perhaps influenced by Alberti, of saturated yellow. It is apparent that
34 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Telesio understood rufus to be a yellowish red and ruber a bluish one. It also
shows that Telesio’s meaning of glaucus has returned to bluish gray.
Hieronymus Cardanus (1501–1576) Italian physician and mathematician
offered a further step in a color value scale.18 The subject of color appears in
more than one of his works, primarily in De gemmis et coloribus (On gems
and colors, 1563). Here Cardanus described an Aristotelian seven-step color
scale as follows: “white, yellow, red, green, wine color, blue, and black.” White
and black are primordial colors, the other five “intermediate.” Additional
mixed colors fall between these primary colors. What is new in Cardanus’s list
is that he assigns numbers of brightness to these colors, thus: “we assume that
white contains a hundred parts of light, scarlet fifty, black nothing.” These
colors fix the beginning, middle, and end of the scale. Yellow is described as
containing 65 to 78 parts of light, green 62, deep green 40, wine color 30, blue
25, and blackish gray 20. This scale is not influenced by concerns of painters
or by issues of color mixing. It is the first-known “quantitative” assessment of
the intrinsic brightness/lightness of object colors.

Sigfrid Aronus Forsius

Significant progress was made by the Finnish mathematician, astronomer and
clergyman Sigfrid Forsius (1560–1624). In 1611, while in Stockholm, he wrote
a manuscript on physics, chapter VII of which is titled “On Vision.” There he
presented two figures in form of circles. The first has sixteen colors ranging in
one direction from white via red to black and in the other direction from white
via blue to black. The red scale has the steps white gold, gold, burnt gold, red,
purple, brown and violet brown; the blue scale dapple, gray, sky blue, blue, blue
green, green, blackish green. Forsius discussed this scale as follows: “Among
the colors there are two prime colors, white and black, from which all others
have their origin. . . . In the middle between these colors, red since ancient
times has been placed on one side and blue on the other one. . . . Gold between
white and red, . . . brown between red and black. . . . Then on the other side
between white and blue is gray. . . . And on the lower part green between blue
and black” (Fig. 2-2a).
With this circle Forsius put classical and medieval ideas on tonal color scales
into graphical form.19 However, he was not satisfied with the result and con-
tinued:

But if you want right to consider the origin and relations of the colors, you should
start from the five principle middle colors which are red, blue, green, gold, and
gray of white and black. And their gradings, they rise either closer to white by
their paleness or to black by their darkness; albeit they are (as above has been
made known) related to one another as previously shown. Because red rises to
white through pale red (pink) and skin color; to black through purple, brown,
violet brown and black brown. Similarly gold relates toward white through pale
gold, wooden and wheat color; to black through burnt gold and blackish brown.
 COLOR ORDER IN THE RENAISSANCE 35

Fig. 2-2a Classical color order is interpreted 67 Forsius.

Equally blue rises to white through sky blue and pale blue, like Dutch cloth; and
to black through dark blue like indigo color that has some brownish to it. So rises
also green toward white through verdigris and pale green; to black through black-
ish green. Gray approaches white by the color of light gray, dapple gray and lime:
to black by mouse gray, black gray and pale black. And this is the correct rela-
tionship of colors that in their number agree with that of the planets as do the
lower colors with the five membranes of the eye, and with the five senses. All this
can be seen from the accompanying figure. (Fig. 2-2b)

Even though some commentators have interpreted the second of Forsius’

circles as a color sphere, with white and black at the poles and four primary
hues: gold (yellow) and blue, as well as red and green, opposing each other,
there is no indication in the text that Forsius had a three-dimensional arrange-
ment in mind. To be interpreted as a sphere with the primary hues in the
proper sequence of a hue circle requires an unusual interpretation of Forsius’
figure: the left semicircle would be paired with the first right curved line to
36 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-2b Forsius’ proposal of tonal scales ending in white and black.

form a circle, and vice versa. How to properly draw a transparent sphere was
well known in the seventeenth century from several sixteenth-century and
earlier books on perspective. On the other hand, Forsius certainly knew that
there are many intermediate hues between his primary hues and with them
many more curved lines. How he would have arranged these is not clear. It is
evident that Forsius’ second figure does not represent a color sphere but is
rather a figure that illustrates four chromatic and an achromatic value scale
beginning and ending at the same points. His manuscript was not published in
book form and thus did not become known outside Sweden.

The Pythagorean ideas on musical harmony as expressed in the tetractys were

expanded by the Roman philosopher and statesman Boethius (ca. 475–525).
Boethius wrote an extensive treatise on ancient music De institutione musica,
influential into the sixteenth century. Figure 2-3, from a twelfth-century man-
uscript of Boethius’ work on music, is the representation of an octave, with the
basic ratio of 6 and 12.20 Additional ratios show fifths (diapente), fourths
(diatesseron), and tone.A diagram of a similar style was used by the fourteenth-
century French mathematician Nicole Oresme (1323–1382) to show the rela-
tionship between various basic polygons.21 The same type of diagram was used
in the 1542 edition of Argyropolus’ commentary on Porphyry’s Isagoge to show
 COLOR ORDER IN THE RENAISSANCE 37

Fig. 2-3 Graphical interpretation of a musical octave (or double diapason) split into dual-tone
cords, after Boethius (fifth c.). From a twelfth-century manuscript of Boethius’ De institutione
musica. This type of diagram was used to represent several kinds of connections in the first
half of the second millennium.

the relationship between the five types of classification of things (see Parkhurst,
1990). The use of such diagrams was thought to show the simplicity of natural
laws as expressed in the classical four areas of study: arithmetic, geometry,
music, and astronomy. It is not surprising that classically schooled thinkers also
used diagrams of this kind to show relationships between colors.

D’Aguilon, Fludd, and Kircher

François d’Aguilon (Aguilonius, 1567–1617) of the Jesuit order in Brussels is
the author of a six volume work on optics Opticorum libri sex (1613), famous
because of its illustrations by Peter Paul Rubens. In regard to colors d’Aguilon
expresses himself incapable of dealing exhaustively with the complexities of
color mixing as practiced by artists. He included a relational diagram, in the
style of Boethius’s diagram, in which he expressed the classical ideas of all
colors generated from white and black. The top portion of the figure can be
interpreted as a semiquantitative tonal description of the three primary hue
colors: yellow close to white and far from black, with blue the opposite and
red halfway between white and black. In the bottom area three important
mixed hues are shown: gold as a mixture of yellow and red, purple of red and
blue, and green of yellow and blue. D’Aguilon’s diagram follows Chalcidius’s
sequence of colors restricting the chromatic primary colors to yellow, red, and
blue, the painter’s primaries (Fig. 2-4).
Robert Fludd (1574–1637), an English mystic, philosopher, and physician,
is known for a re-interpretation of an earlier scale of the colors of urine and
as the first creator of a printed color circle (Fig. 2-5).22 This is a curious, perhaps
mystically influenced, construction. It simply connects the white and black
ends of a seven-color linear arrangement to form a circle.The contents of these
38 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-4 François d’Aguilon’s basic color scale and mixture diagram of 1613. In the upper
portion the tonal scales of white, the full colors, and black are shown, and in the lower section
important secondary colors resulting from mixtures of the primary chromatic colors yellow, red,
and blue.

Fig. 2-5 Color circle of Robert Fludd of 1629, perhaps the first in print.
 COLOR ORDER IN THE RENAISSANCE 39

colors is described as follows: black: no light; blue: more blackness, less light;
green: equality of light and blackness; red: middle between whiteness and
blackness; orange: more redness, less whiteness; yellow: equality of whiteness
and redness; white: no blackness. The Aristotelian number of seven colors and
the designation of red as halfway between black and white are conventional.
What is unusual is the description of colors toward black in terms of light and
toward white in terms of whiteness.
Athanasius Kircher (ca. 1601–1680), German Jesuit with wide interests in
the sciences, author of 44 volumes of writing, professor at the University of
Würzburg and later at the College of Rome, wrote a text on light and color:
Ars magna lucis et umbrae (All there is to know about light and shadow), pub-
lished in 1646. He also considered, in the Greek tradition, colors to be the
result of activity of light and darkness: “Since color is the property of a dark
body or, as some say, a shadowed light, the true offspring of light and shadow,
we must treat thereof. . . .” In the second chapter, On the multitudinous variety
of colors, he used a modified version of d’Aguilon’s color diagram to illustrate
the arrangement of colors (Fig. 2-6). In the cusps formed by semicircles the
mixture colors of the two chromatic colors involved are placed: aureus, viridis,
and purpureus. Below the peaks of the semicircles are located important tonal

Fig. 2-6 Athanasius Kircher’s color diagram in the Boethius style, 1646. Below are his ana-
logues in other qualities and ideas. Tonal scales are shown as by d’Aguilon, with intermediate
tonal colors identified (e.g., cinereus, ash colored, between white and blue). The secondary
mixtures are shown at the intersections of the corresponding tonal arches.
40 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

colors resulting from mixture of the chromatic color with white or black. Thus
subalbum is a mixture of yellow and white while fuscus is a mixture of yellow
and black. The diagram is interesting because of the attached analogies
between colors and various other qualities, properties, and things: light and
shade, taste, the four classical elements, human age, intellect, a scale from God
to plants, and finally, the strings of the Greek lyra.

Francis Glisson’s Color Specification System

Glisson (1597?–1677), an English physician and Regius professor at Cam-
bridge, wrote several books on medicine, among them Tractatus de ventriculo
et intestinis (Treatise of ventricle and intestines), published in London in 1677.
Its chapter IX is surprisingly titled De coloribus pilorum (On the colors of
hair). The main purpose of this chapter seems to have been to present
Glisson’s ideas about a color specification system that could not only be used
to specify the color of hair but of any colored object. Glisson was a supporter
of the idea of five fundamental colors: white, black, yellow, red, and blue. Using
four visually equally spaced scales, a gray scale, and scales from white to full
color of yellow, red, and blue, all object colors could be specified. Glisson used
a novel approach in his scale development. He determined the “equivalent
strength” of his white and black pigments (lead white, carbon black) by making
a mixture that visually fell halfway between the two extremes, that is, a middle
gray. To achieve this, he found that he needed to use a weight ratio of 50 : 1.
Since he wanted to have a scale of 24 steps and the middle gray was to be step
12, he multiplied the weights of his two pigments by 12 to arrive at 600 grains
of white and 12 grains of black. For the next darker step he used 1 grain of
black more and 50 grains of white less, that is, 13 and 550. In this fashion he
completed his scale in both directions (Fig. 2-7).
Glisson also provided a schematical figure of the arrangement of the scales
(Fig. 2-8). He described the gray scale (or, in his terms, blackness scale) as
being straight and the three chromatic scales as being rounded and sideways
from or oblique to (tres . . . scalae obliquae sunt) the gray scale. He also was
aware of intermediate hues but did not indicate how they should be placed.
The midpoints, indicated by marks, are the locations of his pure chromatic
pigments (orpiment, vermilion, azurite or bice) extending in one direction to
white and in the other to black. But Glisson believed that it was not necessary
and a waste of effort to make complete chromatic scales: the half scales from
white to the full pigment color together with the gray scale would suffice for
color specification. Glisson described as an example of a chromatic scale the
redness scale, with 12 steps. He had determined the strength of vermilion
against lead white to be 1 : 20 and used the same methodology as in the gray
scale. Glisson provided an example of color specification for the golden yellow
blossoms of a flower. This color he judged to be equivalent to grade 11 of the
yellowness scale, grade 3 of the redness scale, and grade 2 of the gray scale. If
necessary, he said, colors could be specified to half steps between his grades.
 COLOR ORDER IN THE RENAISSANCE 41

Fig. 2-7 Pigment mixture chart for Francis Glisson’s gray scale. First column: grade of scale;
second column: weight of lead white (in grains); third column: weight of carbon black; fourth
column: reduced pigment ratio.

Fig. 2-8 Glisson’s sketch of the arrangement of tonal scales. The blue, red, and yellow scales
arch over the horizontal gray scale according to their lightness. The vertical dashes denote the
location of the pure pigment.
42 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-9 CIE L* lightness values of reconstructed Glisson gray scale, beginning with white on
the left and ending with black on the right. The scale is only approximately uniform in terms of
CIE lightness. There is a very large step between white and the first mixture.

Glisson’s gray and red scales have been reconstructed using the specified
pigments and weights (Kuehni and Stanziola, 2002). Reflectance curves were
measured and tristimulus values, as well as CIE lightness values L*, calculated.
In Fig. 2-9 the differences in CIELAB (see Chapter 6) L* between the steps
of the gray scale are shown. Near both ends the step sizes increase strongly.
In between, they gradually decline in size toward black. Only extensive visual
scaling would have disclosed this fact. Similar results were obtained for the
red scale with, again, the first and last steps being too large and toward full
red the steps gradually becoming smaller.

Brenner and Waller

On a different front progress was made toward a color atlas. In 1680 the
Swedish painter and archeologist Elias Brenner (1647–1717) published in
Stockholm his Nomenclatura et species colorum (Nomenclature and species
of colors) containing 31 color samples in six groups: white, yellow, red, green,
blue, and black, to assist miniature painters. Six years later, improving on
Brenner to whom he referred, the member of the Royal Society of London
Robert Waller published A Catalogue of simple and mixt Colours, with a Spec-
imen of each Colour prefixt to its proper Name (Waller, 1686). It contains in a
rectangular chart 112 mixed colors (many named in Latin, Greek, French, and
English). The composition of the mixtures is identified by the colorant names
of the column and row headings. As Waller explained: “. . . I have mixt each
 NEWTON’S COLOR DIAGRAM 43

of the Simple Yellows and Reds with each of the simple Blews, and these
Mixtures give most of the mean Colours, viz. Greens, Purples, &c.” (italics in
the original). The column header colorants are Spanish white, azurite, ultra-
marine, smalt, litmus, indigo, ink black. The row header yellow colorants are
(lead white), Naples yellow, gamboge, ochre, orpiment, and umbra, the red col-
orants minium, burnt ochre, vermilion, carmine, red lake, dragon’s blood, red
ochre (carbon black). One-to-one mixtures were made to fill in the intersect-
ing 98 fields. Waller’s expressed idea was to have a scientific systematic
arrangement of colors. He defended mixtures in single weight ratios by indi-
cating that the possible number of mixture ratios was infinite and therefore
not practically doable. In some copies of the printed paper the colorant mix-
tures were dabbed in.

2.4 NEWTON’S COLOR DIAGRAM

It is remarkable that the first explicit, if incomplete, hue circle was the work
of Newton. Isaac Newton (1642–1727), celebrated mathematician and physi-
cist, clarified the composition of white light as a mixture of lights of different
wavelengths.When these are viewed individually they create various hue expe-
riences, the spectral colors. In his early lectures on optics, deposited as a matter
of his job responsibilities as Lucasian professor of mathematics in the library
of Cambridge University (Optica, 1670–1672) Newton was primarily con-
cerned with the refrangeability of light rays and had little to say about color
mixture and logical arrangements of colors. Proposition 3 states: “The colors
white and black together with intermediate ashens or grays are generated from
rays of every sort confusedly mixed.” In proposition 2 Newton described a
scale of eleven hues that he considered “prominent primitives.” He experi-
mented with overlaid mixtures of prismatic lights by appropriately arranging
three prisms and experienced the well-known difficulties in obtaining white
from mixture of two prismatic colors. He found, and expressed in proposition
4, that “[p]rimitive colors can be exhibited by the composition of the neigh-
boring colors on each side of them.”
In his mature reflections on colors, Opticks (1704), Newton introduced a
partly scientifically based color circle and color mixture diagram (Fig. 11 of
Plate II, Part II, Book I) (Fig. 2-10). He described it (in part) as follows:

With the Center O and Radius OD describe a Circle ADF and distinguish its cir-
cumference into seven parts . . . proportional to the seven musical Tones or Inter-
vals of the eight Sounds, contained in an Eight. . . . Let the first part DE represent
a red Colour, the second EF orange, the third FG yellow, the fourth GH green,
the fifth AB blue, the sixth BC indico, and the seventh CD violet, And conceive
that these are all the Colours of uncompounded Light gradually passing into one
another, as they do when made by Prisms. . . . Let p be the center of gravity of
the Arch DE [comparably for q, r, s, t, v, x] and about those centers of gravity let
44 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-10 Newton’s color circle, 1704.

Circles proportional to the number of rays of each Colour in the given mixture
be described. . . . Find the common center of gravity of all those Circles p, q, r, s,
t, v, x. Let that center be Z; and from the center of the Circle ADF, through Z to
the circumference, drawing the right line OY, the place of the point Y in the cir-
cumference shall shew the Colour arising from the composition of all the Colours
in the given mixture, and the line OZ shall be proportional to the fullness or
intenseness of the Colour, that is, to its distance from whiteness. As if Y fall in
the middle between F and G, the compounded Colour shall be the best yellow;
if Y verge from the middle toward F or G, the compounded Colour shall accord-
ingly be a yellow, verging toward orange or green. If Z fall upon the circumfer-
ence the Colour shall be intense and florid in the highest degree; if it fall in the
mid way between the circumference and center it shall be but half so intense,
that is, it shall be such a Colour as would be made by diluting the intensest yellow
with an equal quantity of whiteness; and if it fall upon the center O, the Colour
shall have lost all its intenseness and become a white. . . . if the point Z fall in or
near the line OD, the main ingredient being the red and violet, the Colour com-
pounded shall not be any of the prismatic Colours, but a purple, inclining to red
or violet. . . .

Newton describes with his figure a diagram that is a complete spectral hue
circle as well as an additive mixture diagram. At the same time he uses the
figure to give a specific example of additive color mixture:

. . . suppose a Colour is compounded of these homogeneal Colours, of violet 1

part, of indico 1 part, of blue 2 parts, of green 3 parts, of yellow 5 parts, of orange
 NEWTON’S COLOR DIAGRAM 45

Fig. 2-11 Descartes’s circular diapason, 1650. It may have served as a model for Newton’s
color circle.

6 parts, and of red 10 parts. Proportional to these parts I describe the Circles x,
v, t, s, r, q, p respectively, that is, so that if the Circle x be 1, the Circle s 3. . . .
Then I find Z, the common center of gravity of these Circles, and through Z
drawing the line OY the point Y falls upon the circumference between E and F,
. . . and thence I conclude, that the Colour compounded of these ingredients will
be an orange, verging a little more to red than to yellow. Also I find that OZ is
a little less than one half of OY, and thence I conclude, that this orange hath a
little less than half the fullness or intenseness of an uncompounded orange . . .
this proportion being not of the quantities of mixed orange and white powders,
but of the quantities of the lights reflected from them.23

The form of Newton’s diagram, reflecting his belief in a correspondence

between colors and musical tones, may have been influenced by an illustration
of the relationship between musical ratios in form of a circle (Fig. 2-11) by the
French philosopher and scientist René Descartes (1596–1650).24 In agreement
with the musical scale Newton used the classical number of seven for the basic
colors of the spectrum (the Aristotelian seven colors, however, include white
and black). He was intrigued by the possibility of parallels between musical
harmony and color harmony and pointed out the common reddidshness of
shortest and longest wave colors of the spectrum and compared it with the
similarity of the tones at the beginning and end of an octave (as reported by
Diderot).25 Newton’s work, as is well known, represented a sea change in
thinking about colors, opening furious and extended discussions only settled
some 200 years later.
46 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

2.5 DEVELOPMENT OF THE COLOR CIRCLE

C.B.’s Color Circle

Only four years after the publication of Newton’s Opticks with its spectral
color circle, an anonymous, probably French, author published an image of a
hue circle in color. The Dutch printer van Dole had issued several editions of
a book on painting miniatures, beginning in 1673. The author was only identi-
fied with the initials C.B.26 To date the identity of the author remains unclear.
In 1708 a new edition was published under a new title Traité de la Peinture en
Mignature (Treatise on miniature painting). It contained, for the first time,
images of a seven-hue circle (Fig. 2-12) and of a twelve-hue circle. The author
described the circles as follows: “Here are two figures by which one will be
able to see how the primitive colors, yellow, red, crimson and blue generate
the other colors and which one might call the Encyclopedia of Colors. The first
figure includes the four primitive colors and three composed from them, and
the second includes those same colors with five others which are produced as
much from the primitives as of their composites.” It is evident that the figures

Fig. 2-12 C.B.’s hand-painted color circles, 1708. Left: The seven-color circle; right: the
twelve-color circle. Note that the pigments used for some of the colors have deteriorated. (See
color plate.)
 DEVELOPMENT OF THE COLOR CIRCLE 47

represent pigment-mixing diagrams. In Gage’s view (1993) the author used a

yellowish and a bluish red primary pigment because only with those could he
produce a mix of true red (shown in the twelve-hue circle).
Identity of number of colors between Newton’s circle and that of the first
circle in the Traité may be suggestive, but the anonymous author’s selection
of the seven colors, related to pigments, is noticeably different from Newton’s.
The colors of the seven-hue circle are yellow, orange, fire red, crimson, violet,
blue, and green. Newton’s indigo is missing. A neutral red only makes an
entrance in the twelve-color circle. Here the colors are yellow, golden yellow,
orange, fire red, red, crimson, purple, violet, blue, sea green, green, yellowish
green. We notice that the circle is heavily red-directed: there are seven colors
containing redness and only three colors containing greenness. Yellowish and
bluish colors are balanced with five each. C.B.’s color circle is the first known
explicit and complete (considering also extra spectral purple) circular hue
arrangement. By the year 1800 this book had seen 33 editions and thus was
widely known, at least in artistic circles.

Castel and Schiffermüller

Louis-Bertrand Castel (1688–1757), a French mathematician, Jesuitical priest,
opponent of Newton in matters of color, was deeply interested in a connec-
tion between color and music and hoped to construct a color organ.27 As
described in Projet d’une nouvelle optique (Project of a new optics) of 1739,
by mixing the three basic colors yellow, red, and blue, he constructed a twelve
hue color spiral which he correlated with the tones and half tones of musical
octaves. Surprisingly, and apparently, as Lambert later reported, as a result of
French national preferences, his primary colors were a yellowish red (fire red),
a neutral to reddish yellow (stil de grain, yellow vegetable dye lake), and “true
sky blue.” He derived the twelve colors directly from work with the prism in
a refutation of Newton’s findings of seven basic colors: crimson, red, orange,
golden yellow, yellow, olive, green, sea green, blue, “violant,” “agathe,” and
violet. “Higher-octave” color spiral segments where obtained by adding white
to his original spiral segment, “lower-octave” colors by adding black. He con-
structed scales of twelve steps each of additions of white and black for the
twelve hues of a hue spiral, ending up with a system of 144 colors. Castel hoped
to expand the gradations in all directions and to display the results in a color
room that he envisaged to be covered with colored woven bands or wallpa-
per specimens representing his colors. He attempted to make the gradations
uniform in terms of perceived differences and, on the hue spiral, ended up with
a different number of steps between his three primary colors. Spirals imply
three dimensions and Castel’s color room seems the first implicit idea of a
color space.
Castel’s work was continued and expanded by the Viennese entomologist
and brother Jesuit Ignaz Schiffermüller (1727–1806/9). Schiffermüller did not
share Castel’s enthusiasm for a possible connection between music and color
48 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-13 Continuous color circle of Ignaz Schiffermüller, with twelve classes of colors, 1771.
(See color plate.)

but was interested in a system that could be used, among other things, to
deduce rules of color harmony. This required a physical expression of the
system in considerable detail. Castel’s central color spiral segment was flat-
tened into a color circle. To specify the twelve hues Schiffermüller used more
descriptive hue terms. Again, they were arranged in attempted perceptually
equal steps. Between yellow and red there are two intermediate steps: orange
and fire red; between red and blue there are four steps: crimson red, violet red,
violet blue, and fire blue; between blue and yellow there are three steps: sea
green, green, and olive green. Each of the twelve hues is the key representa-
tive of a “color class,” and Schiffermüller used Roman numerals to identify
them, starting with blue (Fig. 2-13). Next Schiffermüller began to create steps
toward white and black but did not get beyond three examples in the blue
region (Fig. 2-14). His book Versuch eines Farbensystems (Attempt to construct
a color system) was published in Vienna in 1771, thirteen years after Tobias
 DEVELOPMENT OF THE COLOR CIRCLE 49

Fig. 2-14 Schiffermüller’s tonal scales of three blues, from his 1771 work. Some of the col-
orations have deteriorated. (See color plate.)

Mayer presented his three-dimensional system in Göttingen and one year

before Lambert published his book on the subject of a color pyramid (see
below).

Why Color Circles?

One of the questions arising is why, given the linear nature of the visible spec-
trum, hues began to be uniformly represented in a hue circle. The first extant
circular representations of colors appeared in medieval (continuing into the
fifteenth century) diagnostical diagrams of the color of urine.28 Obviously
these were not complete hue circles but may have been influential for circu-
lar depictions of colors of different hues. The circles by Forsius and Fludd do
not represent hue circles. Both of Forsius’ two circles have a central vertical
50 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

axis of the colors of a gray scale, and he placed tonal scales of four primary
hues, red, yellow, green, and blue, on circular segments on either side of the
gray scale. It is not obvious that the circle has a clear meaning. In case of Fludd
it is not unlikely that mystical, perhaps alchemical reasons produced his circle,
which begins with white and ends in black next to it.
Newton’s choice of a circle for his spectral colors may have been influenced
by Descartes’s circular diapason, but it was also based on his knowledge of
desaturation of spectral light by white light, common for all spectral colors.
Thus the radius length is an indicator of the saturation of a color. Newton’s
placing of the spectral hues on the periphery of the circle in proportion to a
musical scale meant that opposite colors were not exactly complementary
colors (if close to them). For this reason, and the varying chromatic strength
of different hues, opposite colors “when mixed in an equal proportion . . . the
Colour compounded of those two shall not be perfectly white, but some faint
anonymous Colour.” Newton also was aware of compounded purple colors
that belong near line D–O in his circle and that, thus, spectral hues and com-
pounded purples can form a closed series. Newton chose a distribution of hues
according to a musical scale on his color circle that, with its common white
center, also represents geometrical parsimony.
C.B.’s color circle, as shown, is a pigment-mixing diagram based on the idea
of yellow, red, and blue as primaries. The circular form, unlikely to have been
influenced by Newton, may have been the result of independent realization
by a painter that, perceptually, a series of hues derived from three primary pig-
ments and produced by mixing neighboring pigments can return upon itself.
As an entomologist Schiffermüller had interest in systematic color arrange-
ment as a means for classifying butterflies and other colored insects. He also
had contact with many artists and was aware of issues in finding harmonious
color combinations, devoting a chapter of his book to this question. In addi-
tion he thought that the time had arrived to create a complete system of colors.
His circle was a first attempt in this direction. Its form was likely derived from
the spiral arrangement of hues of different tonal values of Castel and his pre-
sumed knowledge of Newton’s work and C.B.’s book. By this time the hue
circle had become a convention that made intuitive sense. Many future systems
would be based on it.
The arguments offered so far for the color circle consist of rationalizations
based on likely reasoning and insight available in the seventeenth century.
There is a further and perhaps stronger argument. As will be shown later and
in Chapter 6, a form of data analysis called multidimensional scaling (see
Chapter 3), when applied to spectral colors, results in a geometrical distribu-
tion of data points best fit with a circle. In addition mathematical analysis of
large collections of spectra of colored objects by various methods, but without
consideration of human color vision properties, locates the spectra in spaces
where a series of hues approximately forms a circle. A hue circle therefore is
a pattern that may have become subconsciously apparent and is the result of
one of the strategies of our visual system.
 MAYER AND LAMBERT’S COLOR SOLIDS 51

A full-fledged color order system based on a hue circle was delayed because
of the growing importance of the idea of three primary chromatic colors, pre-
sented as a triangle, that could be used to create all other hues by mixture. The
source of the idea of yellow, red, and blue as three primary colors is unknown,
and it might have developed from dyeing and painting technology. We have
seen these three chromatic primary colors in d’Aguilon’s diagram of 1613.
The same primary chromatic colors are mentioned by Robert Boyle in his
Experiments and Considerations Touching Colours (1664). Using “pigments”
as generic word for colorants he wrote: “. . . there are but few Simple and
Primary Colours (if I may so call them) from whose Various Compositions all
the rest do as it were Result. . . . I have not yet found, that to exhibit this strange
Variety they [painters] need imploy any more than White, and Black, and Red,
and Blew, and Yellow; these five, Variously Compounded, . . . being sufficient
to exhibit a Variety and Number of Colours, such as those that are altogether
Strangers to the Painters Pallets, can hardly imagine. . . . by these simple com-
positions again Compounded among themselves, the Skilfull Painter can
produce what kind of Colour he pleases, and a great many more than we have
yet Names for.” Yellow, red, and blue had been celebrated as key colors in
several works of the contemporary French painter Nicholas Poussin
(1594–1665). Francis Glisson was a supporter of these colors as primary chro-
matic colors. The inventor of four-color printing (yellow, red, blue, and black),
Johann Christoffel Le Blon (1667–1741), published his book on this subject,
Coloritto, in 1725. As a result, in the first explicit steps toward a three-
dimensional color system, the form of a triangular pyramid (tetrahedron) was
used with the primary colors yellow, red, blue, and white at the vertices.

2.6 MAYER AND LAMBERT’S COLOR SOLIDS

Tobias Mayer (1723–1762), self-educated German geographer, astronomer,

and physicist developed, perhaps through his work with map printers, an inter-
est in color and presented in 1758, in Göttingen, a public lecture on the rela-
tionship of colors. A report about the lecture appeared in a local journal. The
lecture was published posthumously in 1775 in a collection of works, Opera
inedita Tobiae Mayeri (Unpublished Works of T. M.) with the title De affini-
tate colorum commentatio (Commentary on the relationship of colors).29 Here
Mayer, for the first time, presented a plan for a systematic color solid, based
on three primary chromatic colors. In principle, the samples representing the
solid were to be spaced to be visually equidistant. The chosen geometric form
of the solid is that of a double tetrahedron. Even though using pigments as a
basis for his work, he also had colored lights in mind: “There are three simple
or basic colors and no more than that, from mixture of which all others can
be generated, which themselves from others, in whatever ratio they might be
mixed, cannot be generated in any way: red, yellow and blue. We see them in
rainbows, but even more distinctly in rays of the sun captured by a glass prism,
52 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-15 Arrangement of 91 colors of the central plane of Tobias Mayer’s double triangular
pyramid, with the color designation scheme based on the three primary colors R, G, and B,
1758.

though there they are accompanied and surrounded by secondary colors.”

Based on experiments with pigments Mayer concluded that there should be
twelve steps between each of the three primary colors as well as (just as
Glisson) twelve steps toward white from the central plane and twelve steps
toward black: “One has to employ such a ratio between colors to be mixed so
that it can be expressed with numbers which are not very large. . . . Neither in
architecture nor music are proportions greater than twelve generally accepted
since such ratios could barely be apprehended by unaided senses.” Mayer rec-
ognized the principle of threshold differences (he may have been the first to
express it relatively unambiguously).
Mayer designated his three primaries with r, g (for gelb, yellow), and b and
indicated their value in the ratio with superscript numbers. In this manner an
equilateral triangle is generated in which there are 91 designated colors (Fig.
2-15). He maintained a constant sum of colorant parts in mixtures with white
and black and on the next level where colors were mixed with one part white
 MAYER AND LAMBERT’S COLOR SOLIDS 53

Fig. 2-16 Schematic representation of Tobias Mayer’s double triangular pyramid color solid.

each, and with eleven divisions per side, obtained 78 colors. Pursuing this
approach he ended up with 364 colors above the middle plane and the same
number below, a total of 819 color samples for his double tetrahedron (Fig. 2-
16), each with its own designation. Mayer did not envisage an explicit central
gray scale made from mixtures of white and black but expected that mixture
of identical parts of the three primary colors (colorants) in the central plane
would produce neutral grays. In the published paper Mayer did not assign spe-
cific pigments to his primary colors. However, in the newspaper report of his
presentation in Göttingen the “Mayer coordinates” of several pigments were
identified, among them, with the identifier 12 (i.e., primary colors), orpiment
(king’s yellow), vermilion, and azurite (Bergblau). (Note that these are the
same pigments as those selected by Glisson.)
It is apparent that Mayer’s proposal was one of theory, applicable also to
mixture of lights. It is known that he made some experiments with pigment
mixture but not to what extent he investigated visually equally spaced scales.
There is no indication that he was fully aware of the nonlinear relationship
between pigment mixture ratio and the resulting visual experiences, as already
Johann Wolfgang von Goethe in his comments indicated (Goethe, 1810).
Mayer’s publisher, Georg Christoph Lichtenberg, added a commentary where
he applied Newton’s center of gravity principle to Mayer’s proposal. He also
printed a reproduction of a color triangle hand painted by Mayer and com-
mented extensively on the difficulties in obtaining a good result. In 1772, three
years before Mayer’s previously unpublished works appeared in print, the
Alsatian mathematician, physicist, and astronomer Johann Heinrich Lambert
(1728–1777) issued a small book under the title Beschreibung einer mit dem
Calauischen Wachse ausgemalten Farbenpyramide wo die Mischung jeder Farbe
aus Weiss und den drey Grundfarben angeordnet, dargelegt und derselben
Berechnung und vielfachen Gebrauch gewiesen wird (Description of a color
pyramid painted with Caulau wax where the mixture of each color from white
54 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

and the three basic colors is arranged, explained, and its calculation and
various uses indicated). He had read the report about Mayer’s lecture and
cited it in his book Photometria (Lambert, 1760). Using Mayer’s proposal as
a basis Lambert developed his own system. He also employed colorant mix-
tures and the same identification scheme. Lambert justified the triangular
scheme from a presumably visually equidistant color circle in which yellow
occupies the 12 o’clock position, the border of red against violet the 4 o’clock
and blue the 8 o’clock positions. He searched among the available pigments
for a yellow that was neither reddish nor greenish and, correspondingly, for a
neutral red and blue and on that basis selected gamboge, carmine (cochineal),
and Berlinerblau (prussian blue, potassium ferric ferrocyanide).30 A condition
Lambert required, beside neutrality of hue, was that his colorants should
approach as nearly as possible the intensity of spectral colors. He commented:
“I leave it undecided if in the future colorants will be found which approach
the spectral colors even closer than the mentioned carmine, gamboge and
Prussian blue.” Realizing from practical experience that the coloristic
strengths of his three colorants was significantly different, he established them
by finding (as Glisson did) middle colors between the three primaries and
reported them as two parts carmine to three parts prussian blue to twelve parts
gamboge. Mixing his primary colorants in a corresponding ratio (12 parts
prussian blue, 12 parts gamboge, and 2 parts carmine) produced a near black
when applied to white paper and Lambert did not see a need for mixtures of
his colorants with black and discarded the lower half of Mayer’s pyramid.
Lambert expressed dissatisfaction with Mayer’s uniform twelve steps. He had
the Prussian court painter Calau experiment to develop originally a six-step,
later a nine-step visually equidistant triangle. Mixing the colorants with a near
water-soluble wax of Calau’s invention and gum resulted in colorations of high
chroma and good stability. Since all three colorants had a high degree of trans-
parency, Lambert believed he could avoid adding a white pigment and instead
used the white of the paper together with increasing dilutions of the colorant
mixtures to achieve the tonal declines toward white. Nevertheless, Lambert
found that he had to produce a total of 67 mixtures to color his pyramid of
108 colors so that the result met his visual criteria. Interestingly, near black
colors are pushed close to basic blue (colors 11, 12, 19, 20; see Fig. 2-17) at the
lowest level rather than at 21, 27, 28 where, based on the gravimetric rule, one
would expect them.
Lambert proposed for the grades along the sides of the triangle a color
naming system that is only moderately intuitive. He used a system borrowed
from the methodology for designating directions around the compass, for
example: 1, blue; 10, blue toward red; 18, bluish reddish blue; 25, bluish red
toward blue; 31, blue red or red blue; 36, red blue toward red; 40, reddish bluish
red; 43, red toward blue; 45, red. For the interior of the triangle he used pri-
marily descriptive color terms such as “chestnut red brown.”
Lambert saw his pyramid as a general color atlas of use to merchants,
for example, to determine if they had fabrics in stock in the desirable colors.
Consumers could use the atlas to decide what color clothing to buy and what
 COLOR CIRCLES FROM HARRIS TO HENRY 55

Fig. 2-17 Identification scheme of primary colors (1, 9, and 45) and color mixes used in the
basis plane of Lambert’s triangular pyramid color solid, 1772.

colors to combine: “Caroline wants to have a dress like Selinda’s. She memo-
rizes the color number from the pyramid and will be sure to have the same
color. Should the color need to be darker or go more in the direction of
another color, this will not pose a problem.” Lambert believed the pyramid to
be of particular use for dyers. If they could find three dyes approximating his
three primary colorants, and after having determined their coloristic strength
relative to Lambert’s primaries, dyers could calculate how much of each dye
to use to achieve a given shade in the pyramid. Other potential users were
artists who, after sketching, for example, flowers in pencil and writing their
color numbers next to them, could reproduce the natural colors in the studio
by referring to the corresponding colors in the pyramid.
Lambert built a wooden pyramidal structure, and included an image of it
in his book (Fig. 2-18). In this display the 108 selected colors could be prop-
erly displayed. For comparison purposes he had Calau paint chips with twelve
common artist’s colorants along the bottom border: naples yellow, king’s
yellow, orpiment, azurite, smalt, indigo, lamp black, sap green, chrysocolla,
verdigris, vermilion, and florentine lake.
Lambert went about the task with a considerable amount of scientific zeal.
Aside from developing an arithmetic of colorant mixture, he explained the
result of black from a mixture of his three primary colorants logically as the
simultaneous prevention of the activity of each of the three colorants by
the other two. His color pyramid is the first attempt to create a geometrical,
physical model of object color experiences achievable with his primary col-
orants, a color solid. The shortcomings of Mayer and Lambert’s proposals will
become apparent as we proceed.

2.7 COLOR CIRCLES FROM HARRIS TO HENRY

Moses Harris
Color circles continued to be proposed in various formats. In 1766 Moses
Harris (1731–1785), an English entomologist and illustrator of insects, dedi-
56 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-18 Depiction of Lambert’s color pyramid, 1772. The lowest level contains the 45 colors
identified in Fig. 2-17. The higher levels contain reduced sets at higher lightness, ending in
white on top of the pyramid. Black is located on the lowest level. The colors displayed on the
front of the model represent well-known artist’s pigments of the time. (See color plate.)

cated a small volume titled The Natural System of Colours to the then presi-
dent of the Royal Academy, the painter Joshua Reynolds. It contains copper-
plates illustrating two color circles, one based on “prismatic” primaries yellow,
red, and blue (whose mixture is illustrated as black, however; see Fig. 2-19)
and one based on “compound” primaries orange, green, and purple. There are
eighteen hues per circle and each hue is illustrated in twenty gradations. The
first circle is said to contain only those colors “shewn by the prism.”The second
one contains “all other colours in nature, not found in the prismatic part.”
Harris provides specific examples of what he means when using color names,
for example: red–vermilion, wild poppy; blue–ultramarine, cornbottle-flower;
green–sap-green, leaves of the lime-tree.
 COLOR CIRCLES FROM HARRIS TO HENRY 57

Fig. 2-19 Prismatic version of Moses Harris’s color circle of 1786. Some deterioration of col-
orants is evident. (See color plate.)

Concerning the twenty gradations of each hue Harris explained: “The

number of colours in this circle are supposed to be eighteen, each of these
being divided into twenty parts or degrees of power, from the deepest or
strongest to the weakest, . . . so that each of the colours in the innermost or
smallest circle contains 20 degrees of power, but each of the outermost but
one.” They then represent tonal scales from the maximal color near the center
of the circle to whitish colors at the periphery. Harris’ circle is the first in which
maximally contrasting colors are systematically opposed: “Suppose it is
58 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

required what colour is most opposite or contrary in hue to red, look directly
opposite to that colour in the system and it will be found to be green . . . of
every colour or teint in the system no one of them contain in their composi-
tions any of the colours of which those on the opposite side are formed. . . .”
Harris remarked on the problems of coloring his system properly. Certain col-
orants (indigo, gamboge, carmine, sap green) are said by him to contain twenty
degrees of power while others do not.

Frisch
A color circle with tonal scales in direction of both black and white was offered
in 1788 by the Prussian painter Johann Christoph Frisch (1738–1815).31 Frisch’s
immediate purpose was to demonstrate tonal colors in much detail as an aid
for painters at a time when subdued coloration was fashionable. Frisch’s color
circle, presented only in outline, consists of 40 concentric rings, beginning with
black in the center. Only two blackish colors are found on the second ring:
blackish red and blue. On ring 3 the range is expanded to eight hues with
reduced blackness: yellow, orange, red, purple, violet, blue, sea green, and leaf
green. On ring 4 the hues swell to sixteen, with further reduced blackess and
on ring 5 to 32, the maximum number. Here the explicit description ends and
rings 6 to 10 were to contain tonal values toward white. Rings 11 to 20 are
even less well described but apparently were to contain further tonal values
ending up in black again at the periphery. Frisch expressed the opinion that
there are more equal-sized hue steps between yellow and blue and between
red and blue than between yellow and red, and he selected the size of his hue
segments accordingly.

Goethe
Because of its importance in aesthetics the color circle of the German poet
and natural philosopher Johann Wolfgang von Goethe (1749–1832), intro-
duced in his Zur Farbenlehre (On the doctrine of colors) of 1810, should be
mentioned here. Goethe’s six-color circle is based on two triangles: the fun-
damental three colors with red on top, and the intermediary colors. The posi-
tion of red results from a process Goethe called Steigerung (intensification) in
which red represents the peak. At the same time it reflects successive contrast:
“To recognize quickly which colors are called forth by this contrast one can
use the illuminated color circle of our tables that is altogether organized
according to natural principles and offers here good service because its oppos-
ing colors are those that are demanding each other in the eye. In this manner
yellow demands violet, orange blue, purple green and vice versa.” Goethe
interpreted his color circle also in regard to principles of color harmony.

Color circles for the primary purpose of demonstrating rules of color harmony
have also been developed by the German painter Matthias Klotz (1748–1821)
in 1816 (see Section 2.10), the English colorant producer and dealer George
 RUNGE’S COLOR SPHERE 59

Field (1777–1854) in 1817, the French chemist Michel-Eugène Chevreul

(1786–1889) in 1839 (see Section 2.12), the German art historian Friedrich
Wilhelm Unger (1810–1876) in 1858, the Austrian physiologist and son of an
artist Ernst Brücke (1819–1892) in 1866, the French physiologist Charles Henry
(1859–1926) in 1888, as well as others.

2.8 THREE PRIMARY COLOR THEORIES

The idea of three fundamental colors received an important boost in 1801 from
the English physicist and physician Thomas Young (1773–1829). He restated
an hypothesis first mentioned by the English glassmaker George Palmer in his
Theory of Colours and Vision (Palmer, 1777).32 Palmer believed that “The
surface of the retina is compounded of particles of three different kinds, anal-
ogous to the three rays of light [necessary to mix all colors]; and each of these
particles is moved by its own ray.” In Young’s mature version the eye is pro-
vided with distinct sets of nervous fibers: “. . . if we seek for the simplest
arrangement, which would enable [the eye] to receive and discriminate the
impressions of the different parts of the spectrum, we may suppose three dis-
tinct sensations only to be excited by the rays of the three principal pure
colours, falling on any given point of the retina, the red, the green, and the
violet, while the rays occupying the intermediate spaces are capable of pro-
ducing mixed sensations, the yellow those which belong to the red and the
green” (Young, 1824). However, Young’s three primary colors were not those
of Boyle, Le Blon, or the dyers and painters, a situation that continued to cause
confusion until resolved by Maxwell and Helmholtz (see glossary entry
primary colors).

In 1809 the English painter and conchologist James Sowerby (1757–1822) pub-
lished A new elucidation of colours, original, prismatic, and material: showing
their concordance in three primitives, yellow, red and blue, and the means of pro-
ducing, measuring, and mixing them: with some observations on the accuracy of
Sir Isaac Newton. In this treatise he attempted a synthesis of ideas about colors
from lights and from colorants. Sowerby experimented with additive color
mixture by placing narrow strips of paintings of his three primary colorants
gamboge, carmine, and prussian blue next to each other and viewing them from
a distance. As a painter he was interested in the effects of adding white and
black pigments to his primary colorants. In this work a two-dimensional color
order system progresses from white in the center to the most intense chromatic
colors and from there toward black on the periphery.

2.9 RUNGE’S COLOR SPHERE

Phillip Otto Runge (1777–1810), German romantic painter and acquaintance

of Goethe, published in his last year Die Farben-Kugel oder Construction des
60 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Verhältnisses aller Mischungen der Farben zueinander, und ihrer vollständigen

Affinität, mit angehängtem Versuch einer Ableitung der Harmonie in den
Zusammenstellungen der Farben (Color sphere or construction of the rela-
tionship of all mixtures of colors, and their complete affinity, with an attached
essay on the derivation of harmony in color compositions), 1810. Runge made
first mention of his color sphere in a letter to Goethe in 1807. The book con-
tains also an essay by Runge’s friend Henrik Steffens Über die Bedeutung der
Farben in der Natur (On the significance of colors in nature).33
Runge distinguished between transparent and opaque colors. The color
sphere was constructed to contain the totality of opaque colors created from
the five elemental colors red, yellow, and blue as well as black and white. The
central color circle has blue on top (in agreement with Castell and
Schiffermüller) green at 4 o’clock and red at 8 o’clock. Of the twelve segments
reddish colors occupy seven, as do yellowish. Bluish colors occupy five and
greenish colors only three segments. The circle is described as a rounding out
of two opposing equilateral triangles (in agreement with Goethe). The first
contains the fundamental colors yellow, red, and blue. The other triangle con-
tains the mixture colors seen as half way between the fundamental colors:
orange, violet and green. When all three colors of a triangle are mixed in equal
amounts, neutral gray results. A gray axis is erected in the center of the circle
so that gray is obtained by mixture of black and white, by mixture of the three
primary chromatic colors, or by mixture of two opposing colors in the hue
circle. The gray axis and the color circle form a double cone that is extended
to the ideal geometric figure of the sphere (Figs. 2-20, 2-21).

Since all three colors, blue, yellow and red stand at the same distance from white
and black, therefore, the center of the color disk in which those three have lost
their individuality through equal activity must be in the same relationship and

Fig. 2-20 Runge’s basic color triangle (R for red, G for yellow, B for blue) combined with the
triangle of secondary colors (orange, violet, green), extended horizontally to a circle and verti-
cally in cone shape toward white and black, 1810.
 RUNGE’S COLOR SPHERE 61

Fig. 2-21 Runge’s double-cone color solid rounded to final spherical form, with white at the
top and black at the bottom of the sphere, 1810.

in the same distance to white as to black as those three. Since both of these points
(the center point between white and black and that of the triangle Blue, Yellow,
Red) mathematically coincide it follows that both must be one and the same
. . . and that from the identical difference a complete indifference results into
which all individual qualities have dissolved. . . . This point, since it is in equal
distance to all five elements, is therefore to be seen as the general central point
of them all. . . . All mixtures that result from the inclination of a point on the
complete color circle toward white or black (a tendency which is common to all
these points) will slowly loose themselves toward white and toward black. . . . as
the differences of all points of the inclination toward white or black from the
central point are radii the points form nothing but circle segments ending in the
poles white and black. . . . Thereby, the complete relationship of all five elements,
through its differences and inclinations, forms a perfect sphere. Its surface con-
tains all five elements and those of their mixtures which are generated in friendly
inclination of their qualities and toward its center all colors of the surface dis-
solve in equal steps into a balanced gray. . . . Every color is placed in its proper
relationship to all pure elements as well as all mixtures and in this manner the
sphere is to be seen as a general table by which he who requires various tables
in his business, can always find the relationship of the totality of all colors. It now
must be evident to the attentive reader that it is not possible to find a plane figure
that is a complete table of all mixtures; the relationship can only be presented
as a solid.

In a hand-colored copperplate figure views of the sphere from the white and
the black poles, a horizontal cross section along the equator and a vertical cross
section through the two poles are given (Fig. 2-22). Unlike Lambert’s, Runge’s
idea was not to offer just a color atlas but to present an idealized theoretical
construct that not only was meant to represent color relations for the painter
but also presumed deeper psychological and mystical relationships gleaned in
part from Goethe. As indicated in the title of the publication, Runge used the
sphere geometry to also develop his ideas about color harmony.
62 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-22 Hand-colored copperplate of Runge’s Farben-Kugel (1810). Views toward the white
and black poles are on top. The equatorial cross section is on bottom left and the polar cross
section on the right. There are four saturation steps between the full color on the surface and
the middle gray in the center of the sphere. (See color plate.)

While representing the most complete color order system so far, Runge’s
proposal still suffers from unresolved issues. The central axis from pole to pole
represents a gray scale, but the full colors of the equatorial circle are of various
lightnesses so that the meaning of the vertical axis is not explicitly defined.
Colorants that in equal mixture produce a neutral gray do not exist, but Runge
did not address the problems materializing when mixing colorants. These and
other issues have only been resolved in the twentieth century.
 THE EARLY DEVELOPMENT OF PSYCHOPHYSICS 63

2.10 THE CYLINDRICAL SYSTEM OF MATTHIAS KLOTZ

Runge was an admirer of Goethe’s Farbenlehre, but another of his country-

men, the Bavarian court and theater painter M. Klotz (1748–1816), was highly
critical of Goethe’s efforts and paid in his country for his views with a lack of
recognition of his own efforts in this field.34 In 1797 he announced in a journal
Aussicht auf eine Farbenlehre (Prospect for a color doctrine), followed in 1806
with Notification concerning a color doctrine, in 1810 with Explanatory noti-
fication, and finally, in 1816 with Gründliche Farbenlehre (a thorough doctrine
of color, published in 300 copies at the expense of the author). Klotz’ book
consists of two sections, one called Chromatik and the other Prismatik, the
former concerned with object colors, the latter with lights. As a painter Klotz
had extensive experience in pigment mixture and believed in the primacy of
yellow, red, and blue. Before Runge, Klotz defined his primary colors in the
sense of unique hues, as later used by Hering. Comparable to Lambert, Klotz
used white paper as his source of whiteness by diluting the pigment disper-
sions appropriately. In this manner he prepared nine visually equidistant
grades (including white) of the three primary pigments, as well as of a gray
scale. The highest degree of saturation was obtained at grade 4 of his scale.
The darker grades were termed “oversaturated” by Klotz. He next mixed
seven intermediary combinations between each of the primaries at grade level
4, resulting in a total of 24 equidistant hues. Klotz believed that in this color
circle complementary colors were placed opposite. Klotz placed neutral
midgray in the center of the circle and placed between each of the 24 full hues
and the central gray three steps of reduced saturation. In this manner he
obtained a color chart of 97 colors, with saturated chromatic colors at the
periphery and neutral gray at the center. Lightness varies considerably in this
chart. Klotz recognized three modification potentials in his system: Buntmod-
ifikation (hue modification), Brechungsmodifikation (saturation modification)
and Hell-Dunkelmodifikation (lightness/darkness modification). Klotz under-
stood the structure of the cylindrical model resulting from these modifications
but refrained from explicitly describing a cylinder model because he could not
see clearly how to bring object colors and light colors in such a system into
agreement.

2.11 THE EARLY DEVELOPMENT OF PSYCHOPHYSICS

Uniformity of spacing of colors in color scales, an atlas, or color solid had been
a clear but difficult to implement goal for Glisson, Castel, Mayer, Lambert,
Runge, and the just mentioned Klotz. It began to be approached from a
different angle as part of general psychological investigations. The idea of a
differential threshold for illumination was introduced by the French mathe-
matician Pierre Bouguer (1698–1758) in the posthumous Traité d’optique sur
la gradation de la lumière (Optical treatise on the gradation of light, 1760). He
64 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

had experimentally determined the intensity a light must have to make a

weaker one disappear. In a particular experiment involving the light of candles
he found that an increment of 1/64 of the intensity of the candlelight had to
be added to it so that the difference was perceptible. Bouguer concluded that
just perceptible differences in brightness were caused by changes in illumina-
tion and represent a nearly constant fraction.35
The term limen was introduced by the German philosopher Johann
Friedrich Herbart (1776–1841). He defined a limen or threshold of conscious-
ness, by arguing that ideas could only emerge above a threshold of activity of
consciousness.

Weber and Fechner

Thresholds of human senses soon began to be investigated quantitatively by
Ernst Heinrich Weber (1795–1878), German anatomist and physiologist.
Weber began with investigating the sense of touch. Later he extended his
inquiries to temperature discrimination and visual discrimination of the length
of lines. Regarding the perception of weight differences Weber stated in 1834:
“For experience has taught us that expert and practiced men feel a disparity
of weight if it is not less than 1/30 of the heavier weight, and perceive the dis-
parity to be the same if, in place of ounces, we put drams.” Regarding length
of lines he found: “The length, therefore, in which the disparity lies, even
though it is two times less on the former instance, is, however, recognized just
as easily, because in both cases the difference between the compared lines
is 1/100 of the longer line.” In 1846 he contributed the chapter Der Tastsinn
und das Gemeingefühl (Sense of touch and common sensibility) to Wagner’s
Handwörterbuch III, part of a multivolume encyclopedia of psychology,
which made his findings widely known. Weber did not state his findings in an
explicit law.
The German physicist and mystic Gustav Theodor Fechner (1801–1887)
began to think about issues of psychophysics before he knew of Weber. In his
mystical work Zend-Avesta (Fechner, 1851) he stated: “If the strength of the
physical activity actually underlying some mental activity at some point in
space and time is measured by its energy b (energy understood in the sense
of mechanics) and if its change, assuming an infinitely small part of time and
space, is named db, then the accompanying change in the intensity of the
mental activity, to be estimated by feeling or in consciousness, is not propor-
tional to the energy change db, but to the relative change db/b. . . .” In his
Elemente der Psychophysik (Elements of psychophysics; Fechner, 1860)
Fechner introduced the term psychophysics and argued that sensation cannot
be measured. All that can be measured are stimuli and the amount of stimuli
that result in a particular sensation or the difference between two sensations.
The smallest difference that results in a noticeable difference is the threshold
difference. Fechner saw the just noticeable difference (JND), expressed in
terms of stimulus, to be the unit of sensation, with the magnitude of sensation
being the sum of JNDs that lead from the absolute threshold to a given sen-
 THE EARLY DEVELOPMENT OF PSYCHOPHYSICS 65

sation. Fechner expressed Weber’s findings, meanwhile known to him, for the
JND increments as follows:

db
dg = K , (2-1)
b

where dg is the difference in sensation, K is a constant, and db is the differ-

ence in stimulus b. Fechner assumed that if this equation is valid for the JND,
it is also valid for small increments of b, the stimulus: “In fact, if one multi-
plies db and b by any number, so long as it is the same number for both, the
proportion remains constant and with it also the sensation difference dg. This
is Weber’s law. If one doubles or triples the value of the variation db without
changing the initial value b, db is also doubled or tripled.” If equation (2-1)
applies, Fechner reasoned, the relation between sensation and stimulus can be
expressed in the simplest form as

g = log b . (2-2)

Fechner saw this formula as a differential formula and integrated it with the
result

g = K (log b - log b), (2-3)

where b represents a threshold; or expressed differently,

Êbˆ (2-4)
g = K log .
Ë b¯

Fechner termed the value b/b the fundamental stimulus value. “The magni-
tude of the sensation g is . . . proportional to the logarithm of the fundamental
stimulus value.” Fechner named this formula the Maasformel (measurement
formula). It expresses the number of JNDs the stimulus is above threshold.
“In the measurement formula one has a general dependent relation between
the size of the fundamental stimulus and the size of the corresponding sensa-
tion. . . . This permits the amount of sensation to be calculated from the rela-
tive amounts of the fundamental stimulus and thus we have a measurement
of sensation.”
Fechner’s work resulted in three types of experimental methods applied in
psychology: just noticeable differences or the method of limits, the method of
right and wrong cases or method of constant stimuli, and the method of
average error. Each of these methods found champions in succeeding years
and is still employed today.
Fechner introduced the idea of “inner” and “outer” psychophysics. Outer
psychophysics considers the relationship between the measurable physical
stimulus and the reported psychological response. Fechner understood inner
psychophysics to mean the relationship between the excitation of nerve
fibers and the mind. Thus he saw the entire process as consisting of physical
66 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

stimulus–excitation–sensation–reported response. He was particularly inter-

ested in where in this chain the cause of the logarithmic relation is found.

Plateau and Delboeuf

In 1872 the Belgian physicist Joseph Antoine Ferdinand Plateau (1801–1883)
read a paper before the Belgian Royal Society: Sur la mesure des sensations
physiques, et sur la loi qui lie l’intensité de ces sensations à l’intensité de la cause
excitante (On the measurement of physical sensations and on the law that con-
nects the intensity of sensations to the intensity of the cause of the excitation).
Plateau asked several painters to halve the perceptual distance between white
and black oil-painted squares and found that that the results were in close
agreement. He also concluded from viewing etchings in different lights that
luminance ratios, and not differences as Fechner required, were applicable and
that the formula connecting luminance with perceived brightness was a power
function.
Objections were raised against Fechner that humans have no intrinsic sense
of the magnitude of a sensation. This objection was countered by the Belgian
physicist Joseph Rémi Léopold Delboeuf (1831–1896) who played an impor-
tant role in early psychophysics. He showed that observers can judge the size
of the interval between two sensations immediately and directly (they can
effectively compare perceptual distance between two different gray samples
against the perceptual distance between two different, say, red samples). Del-
boeuf demonstrated that for sensations to be measured, they only have to be
arranged on a measurable scale, and that absolute magnitude was not
required. This idea of sense-distance was further developed by the American
psychologist Edward Bradford Titchener (see below). Perceptual scaling of
color space since is usually based on Delboeuf’s idea. Delboeuf was a pupil of
the Belgian inventor of statistics Adolphe Quetelet (1796–1874) who was the
first to apply the Laplace and Gauss normal law of error to human data. Thus
Delboeuf was sensitive to the effects of judgment error on the determination
of perceptual increments. Delboeuf also made an elaborate experiment in
which he had observers perceptually halve various distances between gray
samples. His analysis of the results confirmed the applicability of the Weber-
Fechner law. As a result Plateau’s power law fell into obscurity until the 1930s.
Investigation of a visually equidistant eight-step gray scale by the German
psychologist Hermann Ebbinghaus (whom we will encounter again later) in
1887 resulted in measured light intensities not quite representative of the
Weber-Fechner law.

2.12 CHEVREUL’S HEMISPHERIC SYSTEM

Design of a hemispheric color order system was offered in 1839 in his book
De la loi du contrast simultané des couleurs by Michel-Eugène Chevreul
 CHEVREUL’S HEMISPHERIC SYSTEM 67

Fig. 2-23 Chevreul’s concept for a 72-hue circle with twenty grades each. White is in the
center and black at the circumference of the circle, 1839.

(1786–1889), a French chemist and director of the Manufacture Impériale des

Gobelins, the famous tapestry manufacturer. The base plane of Chevreul’s
system consists of a 72-hue color circle based on the primary colors yellow,
red, and blue, separated by equal segments. Twenty grades of lightness of the
corresponding hue are located on radial lines from the white center, ending in
black on the periphery of the circle (Fig. 2-23). In this ladder the full color of
each hue is located at its appropriate level of lightness:36

In each of the scales . . . there is one tone which, when pure, represents in its
purity the colour of the scale to which it belongs: therefore I name it the normal
tone of this scale. . . . If the tone 15 of the Red scale is the normal tone, the
normal tone of the Yellow scale will be a lower number, while the normal tone
of the Blue scale will be of a higher number. This depends on the unequal degree
of brilliancy and luminousness of the colours.
68 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-24 Schematic depiction of the tonal hemisphere raised above Chevreul’s color circle.
Black covers the hemisphere, with white at its origin.

However, later Chevreul placed all full colors on the same grade, a fact that
resulted in considerable confusion about his system by readers of different
book versions. Colors toward the center from the “normal tone” are mixed
with white, and those toward the periphery with black in appropriate amounts
to make the constant hue ladder visually equidistant. Chevreul realized that
there are many other colors of a given hue that he had not represented in his
plane (the plane being in essence a flat representation of the surface of a color
solid). To represent the missing colors Chevreul chose to erect a hemisphere
above the plane (Fig. 2-24). The apex of the hemisphere is occupied by black,
the line between white at the center of the plane and the vertical axis is formed
by a 20-step gray scale. Chevreul then proposed to mix each of the colors of
the base plane in ten steps with increasing amounts of black but comments:
“It is understood that these proportions relate to the effect of the mixtures
upon the eye and not to material quantities of the Red and Black substances.”
In this manner as the angle of the radial lines increases toward 90° the colors
become increasingly blackish. In such a system there would be large visual
steps between the ninth line of colors, nearing the vertical center line, and the
gray scale of the center line, particularly near the core of the hemisphere. This
can be avoided, as Schwarz has pointed out (Schwarz, 1997), if one assumes
that Chevreul had in mind mixtures of the base plane colors with the appro-
priate gray rather than with black so that a smooth transition to the gray scale
would result. Calculations show that in neither case explicit redundancy of
colors occurs, thus negating later criticism of Ostwald and others in this
respect. Assuming the mixtures to be with gray would have the advantage that
all colors in a hemispherical layer would have the same lightness throughout,
a result that Chevreul likely intended. Chevreul’s system was never fully illus-
trated.Ten 72-hue circles, from the full colors with increasing amounts of black,
and twelve constant hue ladders representing the key colors of the base plane
 YELLOW, RED, AND BLUE, FOR A TIME FIRMLY ESTABLISHED AS PRIMARY COLORS 69

were produced under Chevreul’s supervision using paper printing techniques

(Fig. 2-25; Chevreul, 1864).

2.13 DOPPLER’S SPHERE OCTANT

The Austrian mathematician Christian Doppler (1803–1853), discoverer of the

Doppler effect, the change in frequency of energy as a result of relative motion
of the source and the observer, developed an interest in color in connection
with astronomy. He explained changes in the apparent color of stars, often
described by astronomers, as due to the effect that became named after him.
To be able to explain the effect systematically, Doppler required a systematic
color order system.37 He described such a system in a paper with the title
Versuch einer systematischen Classification der Farben (Essay of a systemati-
cal classification of colors) published in the Proceedings of the Royal
Bohemian Society of Sciences in 1847. Placing his three primary prismatic
colors yellow, red, and blue on orthogonal axes with the common origin of
black, he erected an achromatic axis separated by identical angles from the
three chromatic axes. He believed mixture colors of equal intensity between
two chromatic primary colors to be located on circular segments. The result-
ing space has the form of a sphere octant (Fig. 2-26).

At the points R, B, G, M are located pure red, blue, yellow and gray (white) of
intensity AB = AR = AG = AM, the quarter circles BR, BG and GR contain all
binary mixtures of violet, green and orange, therefore not yet mixed with gray
(white). At the points a, b and g neutral violet, orange and green are located.—
Colors tending toward B, R and G are more bluish, reddish, or greenish. . . . The
circle segments MR, MB and MG are the loci for all mixtures of red with gray,
blue with gray and yellow with gray. . . . There is an infinite number of concen-
tric spherical sections [Ma, Mb and Mg] or at least as many as there are grada-
tions of white light from black via gray to the most intense white.

Doppler was the first to develop a three-dimensional color system not based
on object colors. As a mathematician he was able to describe the relationships
between colors in correct mathematical terms (see also Schwarz, 1991).

2.14 YELLOW, RED, AND BLUE, FOR A TIME FIRMLY ESTABLISHED

AS PRIMARY COLORS

The speculations of Palmer, the work of Le Blon and his competitors, and the
choices of Mayer, Lambert, Runge, and others, helped to establish yellow, red,
and blue as the generally accepted primary colors. Young, possibly influenced
by Palmer, also had concluded that there are three primary colors but his
informed considerations made him eventually select red, green, and violet as
the additive primaries.
70 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-25 One of twelve chromatic scales by Chevreul, with local discoloration. The colors
range from white through the full color (grade 11) to black. These scales are located on the
base plane of the hemisphere. (See color plate.)
 YELLOW, RED, AND BLUE, FOR A TIME FIRMLY ESTABLISHED AS PRIMARY COLORS 71

Fig. 2-26 Sphere octant color space of Christian Doppler, 1847. The axes represent the three
primary colors, yellow (G), red (R), and blue (B).

Hayter
In 1826 the English architect and painter Charles Hayter (1761–1835) self-
published a book titled A new practical treatise on the three primitive colours
assumed as a perfect system of rudimentary information. Hayter’s choice of
primary colors were yellow, red, and blue, and he referred to Leonardo,
Newton, and Young as his predecessors. At the same time Hayter understood
that there is a difference “. . . between the properties of such materials as give
their colours to substances suitable to the purposes of art, and the transient
effects of light, which must not be considered as belonging to a system of
mixing colours for the purpose of painting.” Hayter offered six proofs for the
primacy of yellow, red, and blue:

First—That Yellow, Red, and Blue, are entire colours of themselves, and cannot
be produced by the mixture of any other colours. . . . Secondly—Yellow, Red, and
72 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Blue, contain the sole properties of producing all other colours whatsoever, as
to colour . . . ; Thirdly—Because, by mixing proper portions of the Three Primi-
tives together, Black is obtained, providing for every possible degree of shadow.
Fourthly—And every practical degree of light is obtained by diluting any of the
colours . . . by the mixture of white paint. Fifthly—All transient or prismatic
effects can be imitated with the Three Primitive Colours, as permanently con-
sidered, but only to the same degree of compensation as white bears to light.
Sixthly—There are no other materials, in which colour is found, that are pos-
sessed of any of the foregoing perfections.”

Yellow, red, and blue as universal primaries received strong support for a time
from the Scottish physicist David Brewster (1781–1868), best known for his
work on light polarization. In his 1831 book A treatise on optics, Brewster
stated: “Red, yellow, and blue light exist at every point of the solar spectrum.
As a certain position of red, yellow, and blue constitute white light, the colour
of every point of the spectrum may be considered as consisting of the pre-
dominating colour at any point mixed with white light. . . .” He offered spec-
tral curves representing the content of red, yellow, and blue in spectral colors.
Brewster’s theory of three universal primary colors was widely accepted for
some thirty years until Helmholtz and Maxwell demonstrated its errors.

2.15 HELMHOLTZ, GRASSMANN, AND MAXWELL

In 1852 the German physicist Hermann Ludwig Ferdinand von Helmholtz

(1821–1894) clarified the difference between additive and subtractive color
mixture. In the same paper he published a table of mixtures of pairs of five
spectral lights. Only the yellow and blue lights he selected resulted, when
appropriately mixed, in white light. Reading Helmholtz’s account, the German
mathematician Hermann Günther Grassmann (1809–1877), purely on the
basis of logical thinking, demonstrated in 1853 that if Newton’s theory of com-
pound colors is true it must be possible to match any perceived color with
three properly selected color stimuli. These colors must be related to the sen-
sitivities of Palmer’s and Young’s postulated three sensors. Grassmann then
showed that as a result any color of the spectrum must have a complementary
color with the mixture of the two in appropriate ratio adding up to white light.
Grassmann offered a color circle (Fig. 2-27) demonstrating this fact. The color
arrangement in the circle uses the identification of key Fraunhofer lines in the
spectrum (capital letters) for orientation and in this fashion making the circle
for the first time semiquantitative. Grassmann’s circle is an advanced version
of Newtons’s circle. Less than a year after their publication Helmholtz (1854)
experimentally validated Grassmann’s theoretical considerations, thus con-
firming Young’s view.
In 1857 the Scottish physicist James Clerk Maxwell (1831–1879) published
a paper on his color disk mixture experiments in which he described the results
of disk mixtures in an equilateral triangle. Maxwell stated: “From [various]
 HELMHOLTZ, GRASSMANN, AND MAXWELL 73

Fig. 2-27 Grassmann’s color circle, derived from Newton’s circle, 1853. The beginning
and end of the spectrum have been moved to the 12 o’clock position. The sector widths are
identical to Newton’s.

facts I would conclude that every ray of the spectrum is capable of stimulat-
ing all three pure sensations [related to Young’s three sensors], though in dif-
ferent degrees. The curve, therefore, which we have supposed to represent the
spectrum will be entirely within the triangle of colour. All natural or artificial
colours, being compounded of the colours of the spectrum, must lie within this
curve. . . .” In 1855 he had produced a schematic sketch of such an arrange-
ment. The triangle has at its vertices the three primary colors red, violet, and
green. One to one mixtures of these produce intermediate hues carmine, blue,
and yellow. The spectral colors are shown on the circle, with white at its center.
All real colors must fall on or within the circle (Fig. 2-28).
In the first edition of his Handbuch der physiologischen Optik (Treatise on
physiological optics) of 1867 Helmholtz offered estimated spectral sensitivity
curves for the three Youngian sensors (see Fig. 5-3) and a Maxwell-type tri-
angle with the resulting trace of spectral colors (Fig. 2-29). These excitation
curves were later measured first by Maxwell (Fig. 5-4) and later by Helmholtz’s
assistant Artur König (Fig. 5-5). Here Helmholtz also presented a view of the
basis plane, with opposing colors being complementary colors, as well as a view
from the top of his color cone which he compared to Lambert’s tetrahedron.
(In Lambert’s “subtractive” pyramid, black is in the center of the basis plane
and white is at the top, while in Helmholtz’s “additive” cone, white is in the
center of the basis plane and black on top; Figs. 2-30 and 2-31.)
In an 1872 paper on color vision Maxwell stated:
74 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-28 Maxwell’s sketch of the fundamental color triangle based on the primaries red, green,
and violet at the apices, and white in the center. The prismatic and extraspectral purple colors
are located on a circle, with object colors falling within the circle, 1856.

Fig. 2-29 Helmholtz’s estimate of the location of the spectral curve (from the right: red,
yellow (G), green, cyan, indigo, violet, with white at the intersection point) in Maxwell’s primary
triangle, 1867.

Let us . . . suppose the colour sensations measured on some scale of intensity,

and a point found for which the three distances, or coordinates, contain the same
number of feet as the sensations contain degrees of intensity. Then we may
say, by a useful geometrical convention, that the colour is represented, to our
mathematical imagination, by the point so found in the room; and if there are
several colours, represented by several points the chromatic relations of the
colours will be represented by the geometrical relations of the points. This
method of expressing the relations of colors is of great help to the imagination.
 HERING 75

Fig. 2-30 Basis plane of Helmholtz’s color cone with white in the center and the saturated
colors on the periphery, 1867.

Fig. 2-31 View from the top of Helmholtz’s cone, with black on top and tonal colors from black
to the saturated colors.

With these statements Maxwell provided the basis for all future mathematical/
geometrical models of psychological color space.

2.16 HERING

Helmholtz and his theory of color vision was opposed, at times bitterly, by
the German physician and physiologist Karl Ewald Konstantin Hering
(1834–1918) (e.g., see Turner, 1994). In the mid-1870s Hering proposed a seem-
ingly much different idea of color vision based not on three but on four
primary hues: yellow, red, blue, and green, which he called the natural color
76 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

system (Hering, 1920). Hering referred to physiologist Hermann Aubert’s

(1826–1892) statement in his 1865 book Physiologie der Netzhaut (Physiology
of the retina): “If we want to be clear about color sensations, then the words
black, white, red, yellow, green, and blue suffice as main designations and I
may therefore treat them as principal sensations or principal colors. . . .”
Hering developed the concept of antagonistic assimilation-dissimilation
processes to explain brightness perception from black to white. At equilibrium
the two processes produce midgray. Two similar antagonistic or opponent
processes produce four fundamental colors (Urfarben), phenomenologically
simple colors: red versus green, and yellow versus blue. Hering declared the
hues of the four fundamental colors to be the salient features in a color circle.
He used a now classical psychological argument in support:

All hues can be arranged in the circle so that these primary hues divide it into
its four quadrants. All colors in (one) half of this circle are clearly yellowish . . .
all those (in the other) have more or less blueness in common. . . . If one
imagines this color circle halved so that the dividing line passes . . . through two
intermediate colors that are opposite each other, for example, through a violet
and one of the greenish yellows, and if the hues in either half are compared, then
one finds no one chromatic quality common to all these hues. . . . If the dividing
line does not pass through two primary colors we always encounter colors that
have no chromatic property in common with certain other colors of the same
half and thus they do not have the slightest similarity of hue. In this way we
recognize . . . that a rational division of the color circle or grouping of color hues
in terms of their internal relations is possible only by using the four specified
primary colors. (Fig. 2-32)

Adjacent primary hues (but not those opposed) can form hue mixtures. The
hues in their most intense form (C) can be “nuanced” by admixture of white-
ness (W) and/or blackness (S). Hering presented as an example an equilateral
triangle with the full color, white and black at the corners (Fig. 2-33). He devel-
oped arithmetic principles that guide the composition of mixed colors. They
are composed of relative proportions of C, W, and S. Reinheit (purity) of color
can be expressed by C/(C + W + S). Hering was fully aware of the intrinsic
lightness of colors:

If one has a primary blue that is as clear as possible and finds a primary yellow
that one cannot say is either lighter or darker than the blue, then anyone with
good color vision who has even a little practice in color analysis will also observe
that the yellow is less clear than the blue or that it is more or less grayish or
blackish. On the other hand, if he has next to the clearest possible yellow a blue
that does not look decidedly darker than the yellow, then he will see that the
blue is whitish. . . . Moreover, I find a good primary red, that is, the clearest
possible, lighter than the clearest primary green available.

Hering did not construct an explicit color order system, and it is not clear how
he would have done so, that is, how he would have resolved the issue of the
different “brightness” of his primary colors.
 HERING 77

Fig. 2-32 Hering’s diagram illustrating the composition of mixed hue perceptions from the
unique hues located on the main axes. The fractions of blue and red of three mixed hues are
shown bottom right, 1905–1911. (See color plate.)

Fig. 2-33 Schematical view of Hering’s constant hue triangle, with full color r, white w, and
black s on the vertices. Colors along the line parallel to that connecting w and s have constant
chromaticness. On the line r to g are located all colors derived from r by veiling with various
amounts of g, 1905–1911.
78 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Hering was not satisfied with the Weber-Fechner law, which he found not
to apply in his investigations into gray scales. Instead, he proposed a relation-
ship between stimulus intensity and lightness response that equals a hyper-
bolic function. When the luminous reflectance is plotted on a log scale, this
relationship results in an S-shaped function, later found to also apply to cone
response kinetics.
Aubert was the first to recognize that Hering’s opponent color theory does
not have to be contradictory to that of Young and Helmholtz: “. . . if one
strictly distinguishes between the process of excitation and the process of
sensation.”38 This idea was pursued by Helmholtz and further developed by
Schrödinger and Luther (see below). It was the Dutch physiologist Franciscus
Donders (1818–1889) and the German physiologist Johannes von Kries
(1853–1928) who proposed a combination of the two theories in a “zone”
theory, assigning the Young-Helmholtz theory to the cone level and the Hering
theory to a later zone on the visual pathway.39 Despite Donders’s and Kries’s
proposals, the Young-Helmholtz theory of color vision became the leading
paradigm until Hering’s opponent-color theory was resuscitated in the second
half of the twentieth century.

2.17 GEOMETRICAL SYSTEMS OF THE NINETEENTH CENTURY

Cubic Systems
In 1868 William Benson, an English architect, was, in his book Principles of
the science of colors, concisely stated to aid and promote their useful applica-
tion in the decorative arts, the first to propose a cubic system. He placed white
and black on two opposed corners of the tilted cube with yellow, pink, and sea
green on the upper three intermediate corners and red, blue, and green on the
lower three (Fig. 2-34). The center of the cube is occupied by a medium gray.
Variations of a cubic system were proposed by E. A. Hickethier (1940), and
others.

Pyramidal and Cone Systems

A color solid in form of a decagonal pyramid was offered in 1874 by the
German physicist Wilhelm von Bezold (1837–1907), the discoverer of the
Bezold-Brücke color effect. The hues of his decagon are red, orange, yellow,
yellow-green, green, blue-green, blue, violet, and purple. White is located in
the center of the base plane and black at the pyramid’s apex (Fig. 2-35), that
is, it is based on light mixture.
Influenced by Hering, Mayer’s triangular double pyramid, was modified in
1897 by the Austrian psychologist Alois Höfler (1853–1928) into a double
square pyramid. However, the Hering primaries occupy the corners of the
central square of Höfler’s construction (Fig. 2-36).
The issue raised by Hering, but not resolved by him, as to how to reconcile
the lightness of full colors with his arrangement in an equilateral triangle con-
 GEOMETRICAL SYSTEMS OF THE NINETEENTH CENTURY 79

Fig. 2-34 View toward the top, white corner of Benson’s tilted color cube, 1868. The gray
scale is hidden behind the white sphere. (See color plate.)

Fig. 2-35 Bezold’s decagonal color pyramid, with black on top and white at the origin, 1874.
80 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-36 Höfler’s double pyramid based on Hering’s unique hues, with the gray scale on the
vertical axis, 1897.

Fig. 2-37 Tilted double cone of Kirschmann, 1895. Zero and infinite brightness are at the
apices. Colors of constant brightness are on horizontal circular planes.

tinued to be of concern to psychologists. The first attempt to solve it in form

of a tilted double cone was offered in 1895 by the German psychologist and
student of Wilhelm Wundt (see below) August Kirschmann (1860–1932) (Fig.
2-37). Kirschmann was also concerned with the difference between lights and
 GEOMETRICAL SYSTEMS OF THE NINETEENTH CENTURY 81

Fig. 2-38 Ebbinghaus’s tilted double pyramid with rounded corners, 1902. The gray scale is
located on the vertical axis.

object colors. His system was to be applicable to both, and he did not put
“white” and “black” at the two apices of his tilted double cone but the signs
0 and ∞, explaining: “It is an error to set “white” or “black” at the ends of the
axis of a color sphere or a double cone because these expressions do not
correctly designate the extremes of the achromatic series of light sensations
but instead are ideas of a rather complex nature.”
The German psychologist Hermann Ebbinghaus (1850–1909) offered in
1902 a color system in form of a tilted double pyramid. The neutral axis of
Ebbinghaus’s system is vertical, but the central opponent color plane is tilted
along the red–green axis with yellow being closer to white. All surface edges
of the solid are rounded because, as Ebbinghaus argued, the transitions
between colors are fluid (Fig. 2-38). In 1909 the English-American psycholo-
gist Edward Bradford Titchener (1867–1927) offered detailed instructions for
building a square tilted double-pyramid model (Fig. 2-39). Neither Höfler,
Kirschmann, Ebbinghaus, nor Titchener provided quantitative details of the
color arrangements (a painted version of Titchener’s pyramid, colored by the
82 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-39 Titchener’s model of the tilted double pyramid. The rhombic central cross section is
shown on the right, 1909.

bird painter Fuentes, is said to be in existence at Cornell University). Their

aim was to provide conceptual models of the psychological color solid.

2.18 THE NINETEENTH-CENTURY

EXPERIMENTAL PSYCHOLOGISTS

In the footsteps of Fechner and contemporaneously with Maxwell and

Helmholtz experimental psychology began to become a science, with particu-
lar strength in Germany. The sense of vision was of considerable interest.
Experimental psychology claimed Fechner and Helmholtz as its immediate
ancestors. The great man of experimental psychology was Wilhelm Max Wundt
(1832–1920), a prolific researcher and teacher with many well-known pupils.
Under Wundt’s supervision considerable research was conducted involving the
sense of vision. Wundt discussed color order at several occasions. In 1874, in
his Grundzüge der physiologischen Psychologie, he proposed a color sphere,
not unlike Runge’s. But based on the then newly and explicitly investigated
color spectrum, yellowish colors assumed a much smaller portion of his hue
circle. The eight identified main hues are red, yellow, yellow-green, green, blue-
green, blue, violet, and purple. Of the eight sectors reddish and greenish colors
 THE NINETEENTH-CENTURY EXPERIMENTAL PSYCHOLOGISTS 83

Fig. 2-40 Wundt’s cone with black on top and white on the origin, 1893.

occupy four, bluish five, and yellowish colors three. In 1892, in the second
edition of Vorlesungen über die Menschen- und Thierseele (Lectures on the
human and animal soul), Wundt offered a Helmholtzian cone with six basic
hues: red, orange, yellow, green, blue, and violet (Fig. 2-40). White was located
in the center of the base circle and black at the apex of the cone. In Grundriss
der Psychologie (1896), Wundt offered a double cone.

Further Development of Psychophysics

After Weber and Fechner and their immediate followers, major research
activity in psychophysics began to shift to the United States.40 The Polish-
American psychologist Joseph Jastrow worked on the methods of determin-
ing the differential limen and recommended that the general limen be replaced
by the probable error of visual determination (Jastrow, 1888). Before Jastrow
the limen had been defined as the point where 50% of the observers were
above and the other 50% below in their evaluation. Jastrow proposed to move
the point up based on the statistical error of evaluation, specifically to 75%
(the concept of standard deviation was developed only later).
In the same time period the American psychologist James McKeen Cattell
(1860–1944) also worked on psychophysical problems. He experimentally
pursued the idea that general difference thresholds could be determined by
measuring reaction times of observers in psychological tasks. He strengthened
the use of statistical methods for the analysis of psychological measurements
and became the champion of this new approach that stressed the relationship
between the individual and the average observer. In his On the perception of
small differences (together with his colleague, the philosopher G. S. Fullerton)
Cattell criticized Fechner’s law on just noticeable differences as being too
much dependent on introspection and indicated his preference for Fechner’s
84 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

method of constant stimuli, making use of the error law (Fullerton and Cattell,
1892).
Psychophysics faced fundamental criticism by Kries who argued that
numbers applied to sensations are not quantities but merely convenient labels,
an issue not yet fully resolved. In the midtwentieth century the American psy-
chologist S. S. Stevens (1906–1973) developed a theory of sensory magnitude
based on extensive experimental data. He found that such data could usually
be modeled well with power functions. Chapter 3 offers a more detailed view
of psychophysics.

2.19 THE MUNSELL SYSTEM

A decisive step forward toward a systematic arrangement of object colors was

taken by the American artist Albert H. Munsell (1858–1918).41 At the age of
twenty-one he read Ogden N. Rood’s Modern Chromatics (then just pub-
lished), a book that influenced many artists of the period (Rood, 1879). It
offered him a solid foundation in color theory and psychophysics of the time.
After art studies in Paris and Rome Munsell became lecturer in color com-
position and artistic anatomy at the Massachusetts Normal Art School in
Boston. There he began to concern himself with how best to teach systematic
arrangement of colors to his students and how to systematically express pleas-
ing color combinations for artwork. Plotting the colors of one of his pictures
in form of a double spiral in 1898 suggested a sphere to Munsell. He began to
construct small spheres with color fields on the surfac that, when spun, demon-
strated color balance (Fig. 2-41). Early in 1899 Munsell initiated a patent appli-
cation for his color sphere (granted in the year 1900). In the same year he
decided to use the decimal system as a basis of his systematic color arrange-
ment and chose five primary colors: yellow, red, purple, blue, and green. The
central vertical axis of the sphere was to be formed by a 100-step lightness
scale (named “value scale”) and Munsell conceived the then new idea of the
chroma attribute for the colors of equal lightness, extending radially from the
central axis. By applying these and the hue attribute systematically, Munsell
moved away from the tradition of placing the most intense colors of a given
hue on the same central plane. This resulted in the vertical and horizontal
dimensions of his color solid having unambiguous definitions in terms of the
attributes. A portable visual photometer of Munsell’s design provided semi-
quantitative luminous reflectance data for the colors he selected. As early as
1900 Munsell realized that the form of his color solid could not be a sphere
because of the varying intensity of different pigments but would need to be a
spheroid. By 1904 Munsell had made visual uniformity by attribute a key
concept of his systematic color arrangement and had settled on his color ter-
minology. In 1905 he published A Color Notation, a description of the system,
with print reproductions of color charts (Munsell, 1905). In the same year
 THE MUNSELL SYSTEM 85

Fig. 2-41 Artist’s rendition of Munsell’s balanced color sphere, patented in 1900. The sphere
was rotatable to achieve additive color mixture to gray and thereby show the “balance” of the
colors on the sphere. The mirror in the back discloses the blue region of the sphere. (See color
plate.)

Munsell met with Ostwald (see below) and exchanged views on color order,
but both were to pursue their own views on what such a system was to look
like. In 1907 the first version of his Atlas of the Munsell Color System, a port-
folio of eight plates with painted paper samples, was published, followed in
1915 by an extended version with fifteen plates and a total of 880 color samples
(Munsell, 1907, 1915). The two versions offered for the first time detailed inter-
nal views of a color solid.
In 1918, shortly before Munsell’s death, the Munsell Color Company was
formed by Munsell’s son and friends. It produced painted paper chips of
Munsell colors and in 1929 issued the first version of the Munsell Book of
86 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Color, a collection of forty pages, each containing chips of various value and
chroma of one given hue. The system is open ended in that the chips repre-
sent what is possible with available pigments. Identification of the samples is
by hue identifier, value level, and chroma level, for example, 2.5BG5/10
meaning a sample of hue 2.5 blue-green, value 5, and chroma 10. The units of
the three visual scales are not identical in perceived magnitude.
Modifications to the system (the so-called Munsell Renotations) were pro-
posed in 1943 by the Optical Society of America Subcommittee on the Spacing
of Munsell Colors (Newhall, Nickerson, and Judd, 1943) in form of colori-
metric values of the aim colors. They are the basis of the modern system (Fig.
2-42). Expanded editions, based on the Renotations, have been issued in Japan
(Chroma Cosmos 5000 and a condensed version thereof: Chromaton 707), and
the system has also been implemented in the form of textile samples. For more
details and an analysis of the Munsell system, see Chapter 7.

2.20 RIDGWAY’S COLOR ATLAS

In 1912, before the publication of Munsell’s 15 plate color Atlas, the Ameri-
can ornithologist and botanist Robert Ridgway (1850–1925) published Color
standards and color nomenclature, an extensive proposal for what amounts to
a double-cone color order system. It consists of fifty-three plates each with
twenty-seven pigment painted color chips, a total of 1115 chips (excluding mul-
tiple whites and blacks). In several respects Ridgway’s system is a forerunner
of Ostwald’s. It is based on Maxwellian disk mixture. His primary disks were
produced using the brightest then available synthetic-organic dyes. His central
color plane was divided into 71 hue steps, taken in double steps, resulting in
thirty-six full color hues. Ridgway considered six hues to be fundamental: red,
orange, yellow, green, blue and violet. His fundamental red was located at 644
nm (as he had been assured it should be by the then associate physicist of the
U.S. National Bureau of Standards, P. G. Nutting), therefore being a quite yel-
lowish red. His yellow, green, and blue are located near the corresponding gen-
erally accepted unique hues. There are 22 steps containing redness, 17 with
yellowness, 13 with greenness, and 19 with blueness. The hue steps were meant
to be visually equidistant. There are 13 steps between unique red and blue but
only six between unique yellow and unique green. Disk mixtures toward white
and black for all hues were made in three grades each in constant increments
(Fig. 2-43). As Lambert already did Ridgway remarked that the admixtures to
yellow have a much different lightness scaling than those to blue. From the
central neutral gray scale of nine grades to the full color there are grades of
decreasing but always identical admixture of gray. Recent measurements of
an atlas have shown the gray scale to be essentially a Fechner type scale and
the color circle to be less than uniform. (Aside from possible colorant deteri-
oration it is also necessary to keep in mind the limited color measurement
capabilities when Ridgway developed his Atlas.)
 RIDGWAY’S COLOR ATLAS 87

Fig. 2-42 Constant hue page from a modern version of the Munsell Book of Color. The gray
scale is not shown. The chroma scale begins at 1 and continues from 2 at two-grade intervals
to chroma 14. Value grades are shown from 2.5 to 9. Courtesy Gretag-Macbeth Company. (See
color plate.)
Fig. 2-43 Page from Ridgway’s color atlas showing three reddish blue hues lightened from
the central color in three steps toward white and darkened in four steps toward black, 1912.
(See color plate.)
 OSTWALD’S FARBKÖRPER (COLOR SOLID) 89

2.21 OSTWALD’S FARBKÖRPER (COLOR SOLID)

The German chemist and Nobel prize winner Wilhelm Ostwald (1853–1932)
began in 1917 to publish a series of texts on color science, the first of which is
called Mathetische Farbenlehre (Theory of logical ordering of colors). Here
Ostwald described a double-cone color solid of related colors based on addi-
tive disk mixture. He scaled the resulting pychophysical solid using the Weber-
Fechner law to approximate the implied psychological color order. Ostwald
used Hering’s color equation: r + s + w = 1, where r is the amount of the Voll-
farbe (full color, colors of highest intensity), s that of black, and w that of white.
He interpreted the equation in terms of reflectance data in the visible portion
of the spectrum. However, the real colors he used in the disk mixture had, of
necessity, in all cases reflectances different from those idealized. Ostwald
based his system on three fundamental hues: yellow, red, and blue. The hue
circle begins with yellow occupying the 12 o’clock position, proceeding clock-
wise to red, blue, and green (unlike Munsell who placed red at 12 o’clock and
proceeded clockwise toward yellow). Opposing hues are complementary; that
is, when optically mixed they result in an achromatic color. Full colors are
located conventionally on the periphery of the central plane. All colors of a
given hue are placed on an equilateral triangle that is half of a vertical section
through the center of the double cone (Fig. 2-44). The central vertical axis is
a gray scale. As in Hering’s constant hue triangles, lines parallel to the line
connecting the full color and white are lines of equal blackness, lines parallel
to the line connecting full color and black are lines of equal whiteness. Lines
parallel to the line connecting black and white are lines of equal purity. Tonal
colors (tints) toward white are located in the upper half of the double cone
and tonal colors toward black (shades) in the bottom half.
To bring his psychophysical solid into agreement with psychological scaling,
Ostwald applied logarithmic scaling so that 16 grades from black to white and
black to full color, as well as white to full color, resulted in 15 visually equidis-
tant steps according to the Weber-Fechner law. In such an arrangement there
are 120 chromatic samples in a triangle and Ostwald applied a double-letter
system to identify them, in addition to the hue number. The system reveals
slices through the color solid in four different directions: equal hue, equal
whiteness and blackness, and equal purity. In Europe, the abridged Farbkör-
per illustrating on 12 plates 680 samples (24 hues and 8-step gray scale), when
first published in 1919 was a revelation of the multitude of colors, presented
systematically (see Fig. 2-45).
Ostwald’s system was analyzed extensively by Foss, Nickerson, and
Granville in 1944.They demonstrated that, unsurprisingly, it represents a series
of compromises and that it fills only a portion of the available gamut of the
Rösch-MacAdam limits (see Section 2.23). Perceptual differences between
grades in different areas of the solid were found to be of varying size.
Ostwald developed several sets of colorations as atlases of his system, the
most elaborate published as Der Farbenatlas, with 2500 samples (Ostwald,
90 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-44 Color identification scheme for colors in a constant hue plane of Ostwald’s double-
cone color solid (1917). The grey scale is represented on the extreme right by colors aa to pp.
The full color is represented by pa.

1917). An atlas in two different sizes implemented with dyed wool samples
was also offered (Der Wollatlas). Ostwald used his color solid to demonstrate
his rules of color harmony (Ostwald, 1918). Implementations of the Ostwald
system after the Second World War are those by Aemilius Müller, who gave
up complementarity for improved uniform hue spacing (Müller, 1953) and the
Color Harmony Manual of the Container Corp. of America (Jacobson, 1942;
Granville, 1994), issued in three editions.

2.22 GEOMETRICAL SYSTEMS OF THE TWENTIETH CENTURY

After Munsell and Ostwald several new color order systems were proposed,
all built on the ideas of double cones, double pyramids, or cubes. All of these
 RÖSCH-MACADAM COLOR SOLID 91

Fig. 2-45 Vertical cross section through Ostwald’s double-cone color solid illustrating constant
hue colors 1 and 13, with veiling toward white and black. The achromatic scale is at the center.
From Farbkörper, undated. (See color plate.)

systems formed more or less idealized geometrical solids and did not offer any
fundamentally new ideas. Among them are:

• Rounded double cone with tilted central plane from 1920 by the Ameri-
can art educator Arthur Pope (1880–1974). Similar systems were subse-
quently developed in Sweden by Tryggve Johansson (1939) and Sven
Hesselgren (1953), and by Frans Gerritsen (1975).
• Modified double cone by C. Villalobos-Dominguez and J. Villalobos in
1947, developed in Argentina and with an atlas consisting of 7000 samples
produced by halftone printing.
• Double pyramid of 1929 by the American psychologist Edwin G. Boring.
• Tilted cube systems by E. Alfred Hickethier, 1940, issued with 1000
printed samples, and Harald Küppers, 1972, issued with 1400 printed
samples.

2.23 RÖSCH-MACADAM COLOR SOLID

In the early twentieth-century color measurement and specification began

to make significant progress. In 1931 the Commission Internationale de
92 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-46 Projective view of the Rösch-MacAdam color solid containing all object colors as
viewed by the CIE standard observer in standard daylight D65. The axes represent CIE chro-
maticity coordinates x and y and luminous reflectance Y. From Wyszecki and Stiles (1982).

l’Éclairage (CIE, International Commission on Illumination) promulgated the

2° standard observer and standard illuminants A, B, and C, making possible
the specification of colors as seen by the standard observer under a given illu-
minant by three numbers, the tristimulus values X, Y, and Z.Alternately, colors
can be specified in terms of their chromaticity coordinates x and y in the CIE
chromaticity diagram and the luminous reflectance value Y (see Chapter 5).
The Austrian physicist Erwin Schrödinger (1887–1961) developed in 1920 a
theory of optimal color stimuli. It was further developed by Sigfried Rösch in
1928 and by David L. MacAdam in 1935. MacAdam calculated for the 2° stan-
dard observer and illuminants C and A the chromaticity loci of optimal colors
as a function of luminous reflectance Y. The resulting solid represents an
object color space meant to encompass all possible object colors, specifiable
in terms of measured values (Fig. 2-46). Investigations have indicated that this
space is not visually uniform. The CIE later recommended more uniform color
spaces based on transformations of the tristimulus space.
 THE GERMAN DIN6164 SYSTEM 93

Fig. 2-47 Luther-Nyberg color solid for object colors viewed in standard daylight D65, based
on the color moments M1 and M2 and the color weight S. From Wyszecki and Stiles (1982).

2.24 THE LUTHER-NYBERG COLOR SOLID

At about the same time, and employing a different interpretation of a

colorimetric system, the Austrian physiologist Robert Luther (1868–1945) and
the Russian mathematician N. D. Nyberg independently developed a vector
color solid based on what Luther called color moments (Luther, 1927; Nyberg,
1929). The two color moments M1 and M2 are located at right angle and are
defined as M1 = Y - X and M2 = Y - Z. The color weight S, defined as S = X
+ Y + Z, is located perpendicular to the crossing point of the two color
moments. Within the resulting solid all possible object colors are located (Fig.
2-47). The Luther-Nyberg solid is an early linear expression of an opponent
color space where the two color moments are interpretable as greenness-
redness respectively yellowness-blueness system axes. In more modern
systems the color weight is replaced by the luminous reflectance Y. The
Luther-Nyberg solid is located in a psychophysical vector space that is not
visually equidistant.

2.25 THE GERMAN DIN6164 SYSTEM

A comprehensive proposal for a color order system, based on ideas of

Manfred Richter, was made beginning in 1955 in Germany and promulgated
as German standard DIN6164. By 1962 a collection of 590 painted paper chips
was available. In 1984 a glossy set of 1004 color chips was added. The color
solid forms a modified double cone with the distance from the full hue plane
94 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-48 Schematic depiction of the DIN6164 color solid based on attributes hue (T), satu-
ration (S), and darkness degree (D). W denotes the white point. From Richter (1976).

to white much shorter than to black. A schematical view is shown in Fig. 2-48.
The three color attributes are hue number T, saturation degree S, and dark-
ness degree D. Colors of constant hue as defined in the system have constant
dominant wavelength regardless of degree of saturation, and the perceived
hue therefore generally varies slightly along these lines. The 24 simple hues of
the system, modified from the Ostwald system, begin with greenish yellow at
12 o’clock and proceed clockwise toward red. They are intended to be visu-
ally equidistant. Between unique yellow (approximately DIN color 2) and
unique red (DIN 9) there are seven steps, between red and blue (DIN 17)
eight steps, between blue and green (DIN 21) four steps, and from green to
yellow five steps. Six saturation steps were visually scaled at one lightness level
only, and the results were extrapolated to other levels. Colors of equal degree
of saturation are located on roughly elliptical contours in the chromaticity
diagram, not unlike the equal chroma contours of the Munsell Renotations.
The darkness degree D is on a scale of 0 to 10, with twenty steps illustrated.
It is based on a formula recommended by Delboeuf in 1872, with experimen-
tally determined values for the constants:

D = 10 - 6.1723 log (40.7 h + 1), (2-5)

where h is the relative lightness A/A0, with A0 representing the lightness of the
optimal color of the same hue (at the Rösch-MacAdam limit). White has the
value D = 0 and black D = 10.
The DIN system is based on limited visual data (scaling of hue at one level
of saturation, scaling of saturation for eight hues at one level of lightness,
scaling of darkness degree to find the applicable constants in the Delboeuf
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES 95

equation) and inter- and extrapolation to other areas. Three compromises

between psychological results and ease of expression in a psychophysical
system have been made:

1. Expression of hues as constant dominant wavelengths, regardless of

saturation.
2. Straight-line extrapolations of points of equal saturation intervals at one
level of lightness to other levels.
3. Definition of lightness scale based on optimal colors rather than on psy-
chological experiment.

The system was described by K. Witt (1981), K. Richter and Witt (1986), and
G. Derefeldt (1991).

2.26 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES

Intensive study of uniform color scales during and after the Second World War
resulted in new insights about difficulties in representing hue, chroma, and
lightness in a euclidean system. The Munsell Renotations were completed in
1943, but there was an obvious, considerable discrepancy between those data
and the MacAdam color-matching error data of 1942 (see Chapter 6). In 1947
the Optical Society of America formed a research committee to develop as
uniform an atlas with samples of a uniform color solid as possible and a
formula to express the implicit space, usable for calculation of color differ-
ences from colorimetric data. It was headed from 1947 to 1972 by Dean B.
Judd and then by MacAdam. The committee very early on decided to present
its results in form of a crystalline internal structure of color space, thus to
abandon the Munsellian attributes. In the chosen system twelve equal dis-
tances around a center color form a cubo-octahedron (Fig. 2-49). The result is
the Optical Society of America Uniform Color Scales (MacAdam, 1974).
Overlapping cubo-octahedra make it possible to fill color space with this crys-
talline infrastructure. The cubo-octahedra are arranged in a manner that
results in a grid of squares in a given lightness plane, with all differences along
the horizontal and vertical directions of the grid implicitly of equal perceptual
size (Fig. 2-50). Grids at the next higher and next lower lightness increment
level are offset. This arrangement does not result in explicit lines of equal hue
or circles of equal chroma but allows seven different kinds of cleavage planes
that provide new vistas within the color solid (Fig. 2-51). In 1974 an atlas with
558 painted samples was made commercially available and the colorimetric
aim values for these samples have been published (MacAdam, 1978).
The system is open ended, and samples are identified by three rectangular
coordinates L, j, g, where L denotes lightness, j approximately yellowness or
blueness, and g approximately greenness or redness. Formulas have been
96 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-49 A central color (black dot) surrounded by twelve visually equidistant colors located
in the center of spheres and forming a cubo-octahedral structure. From Gerstner (1986).

Fig. 2-50 Square grid constant lightness plane of the L, g, j space with the rotated half
cubo-octahedron placed over it. Grids at the next higher and lower lightness levels are
offset by one half chromatic step, as illustrated by the portion of the cubo-octahedron. From
Gerstner (1986).

developed that attempt to represent an accurate psychophysical model of the

psychological space (see Chapter 6). They include adjustments of the lightness
scale for the crispening effect and for the Helmholtz-Kohlrausch effect (see
Sections 5.8 and 8.2 for discussions of lightness and chromatic crispening and
Section 5.8 for the Helmholtz-Kohlrausch effect). For more details and an
analysis of these scales, see Chapter 7.

2.27 SWEDISH NATURAL COLOR SYSTEM

A system called, after Hering, Natural Color System was developed in Sweden
in the 1970s. It traces its sources to Hering and Ostwald, and more recently to
 SWEDISH NATURAL COLOR SYSTEM 97

Fig. 2-51 View of MacAdam’s model of the OSA-UCS color solid illustrating the existence of
several cleavage planes. See text for more detail. Slide courtesy D. L. MacAdam. (See color
plate.)

efforts by Johansson (1949) and Hesselgren (1954). The fundamental idea was
that any color normal observer is able to determine in any object color the
“content” of one or two fundamental full colors and of black and white. The
three, respectively four, percentages always add up to 100%. In a manner com-
parable to Ostwald’s, the Natural Color System is represented by a double
cone. A forty-hue circle of full colors based on experimentally determined
unique hues is located at the edge of the double cone (Fig. 2-52). The central
axis is represented by a gray scale with ten steps, with white on top. Mixtures
of full colors with white or black are located on the surface of the double cone.
Mixtures of full colors with white and black are located in the interior. Colors
of equal hue are located on triangles bounded by white, black, and the full
color (Fig. 2-53). Colors are identified by hue number, NCS chromaticness c,
and NCS blackness s. An atlas representing the system has been developed,
containing 1750 matt color chips arranged in triangles on forty equal hue
planes. Every other plane illustrates colors from c = 10 while those between
98 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

Fig. 2-52 Hue circle of the NCS color solid with hue designation scheme, identifying ten hue
steps between unique hues. Courtesy NCS.

begin at c = 40. Colorimetric values based on the 2° standard observer of the

aim colors and typical color chips have been provided in a table (Swedish Stan-
dard, 1982). For more details and an analysis of the NCS system, see
Chapter 7.

2.28 UNIVERSAL COLOR LANGUAGE

In 1931 the then just formed (American) Inter-Society Color Council (ISCC)
began work under Godlove on a standardized method (in English) to desig-
nate colors. An initial list was issued in 1939, and Kelly and Judd issued in 1955
NBS Circular 553 The ISCC-NBS Method of Designating Colors and a Dic-
tionary of Color Names. In 1976 the NBS Special Publication 440 Color: Uni-
versal Language and Dictionary of Names was issued (Kelly and Judd, 1976).
The Munsell color solid was subdivided into regions that can be designated by
common names. The system is applicable to opaque, translucent, and trans-
parent colored materials. Twenty-nine major color regions contain a total of
267 subregions around centroid colors. Among the twenty-six major regions
are those of the primary colors black, white, yellow, red, blue, and green, as
well as secondary and tertiary colors such as orange, brown, olive, purplish
pink, and violet. The subregions are identified by modifiers, such as vivid,
strong, deep, dark, light, and grayish. A supplement contains on eighteen pages
a sample of the centroid color of the 267 subregions, with its ISCC-NBS name
and Munsell Renotation.
In the 1976 publication Kelly describes six levels of precision of designat-
 UNIVERSAL COLOR LANGUAGE 99

Fig. 2-53 NCS constant hue triangle of hue Y90R with full color C, white W, and black
S. Colors of constant blackness s are located on lines parallel to W–C, colors of constant
chromaticness c fall on lines parallel to W–S. Courtesy NCS. (See color plate.)

ing colors. Level 1 consists of thirteen generic color names: white, gray and
black, yellow, orange, brown, red, pink, purple, blue, green, olive, yellow green.
On level 2 sixteen intermediate hue names are added, such as reddish brown,
bluish green, and purplish pink. Level 3 contains the 267 subregions. On level
4 the space is further subdivided into more than 1000 colors, as in the Munsell
Book of Color. Some 100,000 colors can be specified on level 5 by visually
interpolating between Munsell chips, such as a color specified as 7.8PB
100 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

2.2/12.5. On the final level colors are specified by standard colorimetric

data, such as CIE x, y, Y values. Here the number of definable colors is in the
millions.

2.29 COLOR MIXING SPACES

In computer and color reproduction technology of the late twentieth century

several types of color space are in routine use. In cathode ray tube (CRT) dis-
plays, color television, and most computer video displays, color stimuli are gen-
erated with three different types of phosphors that can be activated with
electron beams. The colors resulting from such activation typically are orange-
red, leaf green, and violet, additive color primaries. A large variety of color
experiences can be created as a result of their mixture. These stimuli are rep-
resented in the RGB color cube. The abbreviations R, G, B designate red,
green, and blue, loose verbal designations for additive color primaries used.
The cube resembles the Benson cube (see Section 2.15). Colors can be iden-
tified in terms of R, G, B units of activation. In the RGB version the standard
values of the three components range from 0 to 255. As the cube is rotated so
that white and black fall on the vertical axis, a version of a polar coordinate
system is imitated and the system is termed HSB (for hue, saturation, bright-
ness) space. In this model hue is expressed as hue angle in degrees, saturation
as a percentage from 0% at the achromatic point (gray) to 100% at full satu-
ration, and brightness as a percentage from 0% at black to 100% at white.
Achromatic colors have identical values of the three components, chromatic
colors have varying values. The phosphors used define the gamut (the
maximum chromatic range) of colors that can be achieved. The HSB and RGB
spaces depend on the exact definitions of the primary colors used in their cre-
ation. They are regular spaces but not uniform as defined in Chapter 1.
Software programs such as Adobe Photoshop® display slices through the
HSB space with a continuous gray scale on the left-hand side and the full
colors at different lightness on the right. The plane is filled with intermediate
colors (Fig. 2-54). Because of the additive mixture the hue of such a plane is
not uniform but changes slightly as a function of saturation and brightness.
Another space used in design software is the CMYK space.The letters stand
for the subtractive primaries of four-color printing: cyan, magenta, yellow, and
black (K to avoid confusion with B). In printing processes the primary colors
are defined in terms of standard printing inks. For CRT simulation (e.g., in
Adobe’s Illustrator® software) CMYK is defined in terms of individual per-
centages of the four components. A white color has zero values for all com-
ponents. The gray scale differs in percentage of K. A chromatic color may have
percentage values in all four categories.The gamut of CMYK is usually smaller
than the gamut for RGB because of the limited chroma of printing primaries.
In programs like Illustrator® and Photoshop® colors can be defined simul-
 SPECTRAL SPACES 101

Fig. 2-54 Color Picker screen from Adobe® Photoshop showing specification of a given
reddish blue color in four systems: HSB, Lab (=L*, a*, b*), RGB, and CMYK. (See color plate.)

taneously in terms of HSB, RGB, CMYK, and CIELAB L*a*b*, and they
allow input values from any of these systems to display (with a degree of accu-
racy that depends on many factors) the corresponding color. There are
complex issues of color management in reproducibility of such CRT colors in
different media (color transparency or opaque color photos, printing on dif-
ferent types of printers, from computer printers to commercial printers).

2.30 SPECTRAL SPACES

Spectral functions can be placed in regular spaces without consideration of

human observers. Such spaces have use in machine vision, scanners, and
copiers and other applications. Several methods have been applied for this
purpose. Among these is principal component analysis (PCA), first used in
connection with colors for a subset of Munsell colors by J. Cohen (1964).
Another is through training of computerized neural networks. In PCA basis
functions are calculated describing the spectral functions with a degree of
accuracy that depends on the number of functions. If the number of basis func-
tions is three, they form an orthogonal space in which each spectrum plots as
102 HISTORICAL DEVELOPMENT OF COLOR ORDER SYSTEMS

a point. For an extensive set of samples, such as those of the Munsell system
or of images of natural scenes, about 90% of the variation in the spectral func-
tions is explained by three basis functions (Lenz et al., 1996; Wachtler et al.,
2001). There is an all-positive nearly horizontal function that can be seen as
being the spectral equivalent of a lightness function. The next two functions
have positive and negative lobes and can be seen as rough approximations
of opponent color functions (see Section 6.18). They extract information in
spectral functions more efficiently than functions describing the spectral
sensitivities of human cones. On the other hand, while both kinds of
functions place the reflectances of a Munsell hue circle into ordinal order,
(with some exceptions) only the cone and color-matching functions place them
approximately in interval order in regard both to hue difference and chroma.
There are several mathematical treatments that can be used for PCA and
similar operations. Some of these have found use in color constancy and color
appearance modeling (Maloney, 1999). However, for the reasons given and
others it is not justified to call spectral spaces based on reflectance functions
color spaces.

This review of the historical development of thought on color order and of

color-ordering systems has concentrated, with few exceptions, on psychologi-
cal systems that have often found a physical expression in form of atlases.
When comparing such atlases, the samples are found to differ. There is a con-
siderable literature on the results of comparisons among the Munsell, NCS,
DIN, and OSA-UCS systems (e.g., see Derefeldt, 1991). The results have been
summed up by the philosopher of color C. L. Hardin (1988): “. . . the issue is
not that there cannot be a consistent scheme of representing phenomenal
color, for there can. And, indeed, there is reason to suppose that the various
representations can be mapped into each other (though the mapping relations
will often be quite complex).” When doing such mapping, one should
distinguish between systems that aim to be perceptually uniform (e.g., Munsell
and OSA-UCS) and other systems that are regular in some way (e.g.,
NCS). Accordingly we may expect common principles behind the first
group but significant discrepancies between the two groups, as will be seen in
Chapter 7.
In discussion of the Munsell system in Chapter 1 we have seen early scaling
experiments that resulted in significantly different ratios of hue differences to
value and chroma differences depending on the size of the difference. This
points to an apparent dependence of color space on the size of the differences
selected to represent the space, a matter discussed in detail in Chapter 6.
As this chapter indicates, development of a uniform color space has been
a struggle over more than two centuries, a struggle that has not yet been suc-
cessful. A modern view for the reason is that our color perceptions are empir-
ically determined and thus not in a relationship to stimuli that is expressible
by simple logical rules (Purves and Lotto, 2002) or that there are no explicit
neural correlates for what we perceive (Dennett, 1991; O’Regan and Noë,
 SPECTRAL SPACES 103

2001). Under conditions of extensive relativity (average observer pool, con-

stant defined lighting, and surround conditions) it may be possible to reduce
the visual situation to one in which no ambiguities are present, and empiri-
cism may be limited to a constant set for the observer pool. The modest degree
of agreement in different attempts at a visually uniform space in the last
century perhaps gives an indication that this is possible. Additional experi-
mental work is required to establish what is feasible. The value of such a space
would be limited to the specific conditions under which it was determined. For
many applications this can nevertheless be useful.
 Chapter 3
Psychophysics

Seeing and hearing have a comparable function: transforming types of ubiq-

uitous physical energy into a format that brains can use in support of the
organisms they are a part of. Tasting and smelling rely on the interaction of
organic or inorganic chemicals in the environment with special chemical
sensors in the corresponding organs. Other, specialized sensations, such as
certain feelings of pain or pleasure, presumably result from the generation of
electrochemical signals entirely within the body. Psychophysical methods are
empirical attempts to discover the connection between stimuli and the result-
ing sensory experiences by measurement of reported percepts or performance.
By categorizing experiences and analyzing the connection between stimuli
intensities and the resulting experiences, psychophysics attempts to discover
how sensory psychology operates and to help establish linking propositions
between neurobiology and psychology.1 One might expect sensations to be
linearly related to stimuli, but this is rarely the case as will become apparent.
The mathematical underpinnings of psychophysics are related to measure-
ment theory, a branch of applied mathematics. A postulate of measurement
theory is that measurements are not the same as the attribute being measured.
By measuring the difference between two colors, we have not revealed any-
thing about the nature of the two colors. W. S. Sarle (1995) defines measure-
ment of an attribute of a set of things as “the process of assigning numbers or
other symbols to the things in such a way that relationships of the numbers or
symbols reflect relationships of the attribute being measured. A particular way
of assigning numbers or symbols to measure something is called a scale of
measurement.” The mathematics of measurement are complex and have been

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

104
 FUNDAMENTS OF PSYCHOPHYSICS 105

described in the three-volume work Foundations of Measurement (Krantz et

al., 1971, 1989, 1990).

3.1 FUNDAMENTS OF PSYCHOPHYSICS

The problem at the heart of psychophysics is lack of understanding, and of a

theory about, how our feelings and experiences are generated. It is what the
philosopher D. J. Chalmers referred to as “the hard problem” of consciousness
(Chalmers, 1996). Given the lack of such knowledge attempts to create a
fundamental theory of psychophysics are speculative. Fechner faced critique
from his contemporary Kries that sensation magnitude is the result of hidden
processes (that sensation is cognitively impenetrable) and that, as a result,
numbers applied to sensation are not numerical in the mathematical sense but
merely convenient labels. Numbers may erroneously suggest that quantitative
measurement in the sense of physical measurement has taken place, implying
a precision that is not there.
Psychophysics has not answered this attack in a fundamental way. It needs
to be seen as an essentially empirical, pragmatic attempt to connect intensity
of stimulation with magnitude of sensation or, rather, perception. Measuring
or scaling psychological magnitudes or differences involves making judgments.
Judgment is defined in the relevant meaning as “the process of forming an
opinion or evaluation by discerning and comparing” (Webster’s Collegiate
Dictionary, 10th ed.). The processes by which judgments are formed and their
neurological expression are unknown. If S is the experienced sensation and R
the judged numerical response to the stimulus intensity I, there is an implicit
psychophysical transformation F1 as follows:

S = F1 (I ). (3-1)

On the other hand, the response of observers, R, is related to intensity by the

judgment function, F2:

R = F2 (I ). (3-2)

The two transformation functions F1 and F2 are normally conflated to arrive at

the result

R = F (I ). (3-3)

Without knowing the values of either component it is impossible to determine

the other (Marks and Algom, 1998). Equation (3-3) is a mathematical expres-
sion of the assumption underlying psychophysics.
A recent attempt, among several, to develop a fundamental psychophysi-
cal theory has been made by K. H. Norwich (1993). He based it on informa-
106 PSYCHOPHYSICS

tion theory. More intense stimuli have greater information content than less
intense ones. According to Norwich sensation provides a measure of the infor-
mation content in the stimuli:

S = kH , (3-4)

where S is a perceptual variable (the sensation magnitude), k is a positive con-

stant, and H is the stimulus information available for conversion. S is taken to
be the result of several separate stimuli, and H is calculated as a function of
the probability of each of the stimuli involved. While the stimuli may in some
cases be easy to define in a physical sense, the sensors (e.g., the cones of human
vision) are believed to continually undergo changes at the microscopic level.
The Weber-Fechner law as well as Steven’s power law have been shown to be
special cases of Norwich’s law, with the Weber-Fechner law presumably apply-
ing at higher levels of information and the power law at lower. By itself
Norwich’s law is of course not an explanation of the “hard problem.”
Psychophysical measurements can be made at different points in the seeming
continuum of experience: (1) absolute threshold, (2) difference threshold (just
noticeable differences), and (3) scaling of the full continuum. Absolute and
difference thresholds have classically been determined by Fechner’s methods,
described below. In the 1960s the signal detection theory was developed (Green
and Swets, 1966), and it breaks Fechner’s idea of a discrete sensory threshold
into two components: (1) a neurophysiologically determined basic component
(usually called discriminability, d¢, and (2) a cognitively determined decision
process (response bias, b). As a result there are four response types possible:
hits, misses, false alarms, and correct rejections.
Fechner believed that sensory scales can be derived on basis of discrimi-
nation and that just noticeable differences expressed in terms of stimulus
increments or decrements can add up to sensory magnitudes. Stevens, on the
other hand, postulated that sensory magnitudes can be directly assessed.
According to him measurement is the assignment of numbers to objects
according to certain rules and results in empirical relations. There are experi-
mental data in support of either approach. Psychophysicists today seem to lean
toward the idea that the fundamental process involves taking differences.
Establishing sensory scales may be said to involve adjusting distances between
stimuli to match an internal or external standard, or to make numerical esti-
mates of sensory differences.
Psychophysical judgments are known to be subject to various contextual
effects. For example, the results of the scale halving method are known to
depend on if the stimulus samples are presented in bottom-up or top-down
sequence (hysteresis effect). For color scaling as for many other types of per-
ception, the level of adaptation strongly affects the resulting scale. As will be
seen in Chapters 4 and 5, gray scales depend strongly on the lightness level of
the surround.
Stimulus sequence has also been shown to affect scaling results. Among the
 FUNDAMENTS OF PSYCHOPHYSICS 107

possible reasons are changes in the adaptation level based on the previously
seen stimulus. The range of stimuli displayed and the level of stimulus are also
known to affect the result, particularly in case of magnitude estimates. Among
midgets and basketball players the meaning of small, medium, and large
involves absolute values that are much different.
Context effects are well known. How the lightness of a surface is assessed
depends not only on the adaptation level but also on other contextual infor-
mation such as is it seen in shadow or in direct illumination. Similar effects
apply to chromatic surfaces.
Despite its lack of a solid foundation psychophysics persists because it has
considerable pragmatic value. The continuing interest in psychophysics may
also be due to what L. E. Marks called “the metaphorical imperative” (1978).
Accordingly we have a need to express our inner world in terms of metaphors
that are more easily comprehended by other humans. Scaling and magnitude
determinations of perceptions can be seen as an expression of the metaphori-
cal imperative.
When viewed from the perspective of Fechner’s outer and inner psy-
chophysics, mentioned in Chapter 2, the conventional psychophysical enter-
prise can be represented in the schematic sketch of Fig. 3-1. The relationship
between stimulus and conscious experience represents classical psychophysics.
The relationship between nerve excitation and inner sensation is in the domain
of neurophysiology and as such subject to process variability that can be

Fig. 3-1 Conventional psychophysical enterprise as viewed from the perspective of Fechner’s
inner and outer psychophysics (modified from Murray, 1993). Outer psychophysics attempts to
connect the stimulus magnitude with the response magnitude. Inner psychophysics attempts
to connect nerve excitation with nonconscious “sensation.” It is quantified by recording from
visual system cells. Statistical variability and signal detection are taken to be aspects of inner
psychophysics.
108 PSYCHOPHYSICS

treated in different ways, as we have seen. The classical view of psychophysics

is that humans react to stimuli. The modern view is that people act on stimuli
by interpreting them in a way that contributes toward achieving the goals of
the individual.

3.2 CATEGORIES

Categorization is a fundamental trait of humans and perhaps of all animals.

To distinguish between edible and poisonous plants, meaning to put them into
categories, was an early and imperative task for plant eaters, for example.
An important human method of categorization is naming. The process by
which we establish categories and sort stimuli into the established categories
is clearly complex and at this time largely unknown. Several different theories
have been established and vie for recognition.
Stimulus dimensions available for classification are either taken to be con-
tinuous (e.g., color) or discrete (four-sided vs. three-sided figures). The struc-
ture of categories is either overlapping (hue and chroma) or nonoverlapping
(color and shape). In the former case perfect categorization is difficult or
impossible while in the latter it is likely. The accuracy of response in catego-
rization experiments is fundamentally either deterministic, meaning equal
stimuli result in an equal categorization, or probabilistic: the response of the
observer is always based on a (more or less informed) guess. It is obvious that
already the stimulus is usually probabilistic. In case of visual stimuli, for
example, light is reflected probabilistically off the surface of the object as well
as off the eye’s cornea, and once the remaining stimulus enters the eye, there
is probabilistic activity at all levels of the vision system. Most categorization
theories therefore treat categorization data in terms of variability.
Among the theories of category access (what process shapes our category
decisions) are those of the necessary and sufficient condition (NSC), the
prototype theory, and, related to it, the exemplar theory. According to NSC,
categories are described by a series of necessary and sufficient conditions, and
the observer tests the sample to see of it meets these conditions. In some sit-
uations (e.g., separating squares from triangles), NSC is clearly applicable, but
in many others, it is not (e.g., it is difficult to define the necessary and suffi-
cient conditions for a letter symbol to represent an a). The prototype theory
proposes a (kind of Platonic) prototype for a category whereby objects are
classified by their resemblance to the prototype. It implies that we only have
the prototype stored in memory and that determining resemblance to it is a
cognitive task. It simply delegates the critical categorization to another
process. To get out of this difficulty, the exemplar theory was proposed.
Accordingly we have all examples of past experiences of the category item
stored in memory and compare new examples to the stored ones to catego-
rize them. It seems doubtful, however, that observers need to have seen all
reddish colors before they can categorize them as such, for example.
 DIFFERENCES VERSUS MAGNITUDES 109

Visual classification is believed to take data from the visual area at the back
of the brain to the inferior temporal cortex of the brain located near the
temples. This region appears to analyze the data for certain features. It
exchanges data with the prefrontal cortex area in the front of the brain an area
that may contain category border codes. Categorization is achieved jointly
between the latter two areas (Hasegawa and Miyashita, 2002).
In regard to color classification the fundamental question is what process
results in setting classification boundaries. If given a thousand randomly dif-
ferent Munsell color chips and asked to categorize them, how would we do it
if we had not been exposed to theories and examples before? An obvious
attribute for categorization is hue. But it has taken until the seventeenth
century to recognize that hue forms a closed circular continuum. Newton has
separated spectral colors into seven categories and saw that bluish red and
purple colors, not found in the spectrum, can close the spectral array into a
circle. Rational classification of hued colors according to whiteness and black-
ness or lightness and chroma only took place in the nineteenth and early twen-
tieth centuries.
There is the question of the cause of human color categorization in lan-
guage. As briefly mentioned in Chapter 2, there is a theory by Berlin and Kay
that basic color terminology in human languages has followed the same
pattern. They identified eleven basic color terms, aside from the six Hering
fundamental colors: brown, purple, pink, orange, and gray. The list is notable
for its idiosyncrasy: all four chromatic colors are in the yellow-red region of a
color circle. In addition the English term pink had in the sixteenth and
seventeenth centuries the words yellow or brown attached to it (Merrifield,
1967), and the color of English hunting jackets is called pink. There are no
terms for yellow-green, blue-green, and their dark, desaturated versions olive
and navy. While there should have been a good number of such colors in
natural surroundings in many environments, it has taken systematic catego-
rization of the hue circle to recognize these colors as categories. In C.D.’s
extended color circle they are listed as yellowish green and sea blue (see Fig.
2-12). A convincing theory of color categorization that explains the historical
development of basic color terms seems to be a long way off.

3.3 DIFFERENCES VERSUS MAGNITUDES

The Weber-Fechner law assumes that just noticeable differences can be con-
sidered units of perception. It states that the increment required to result in a
just noticeable perceived difference is a constant percentage or fraction of the
stimulus magnitude (see Chapter 2). The Weber-Fechner law has been found
to be context sensitive.Typical experimental Weber fractions are given in Table
3-1 and an example of the Weber-Fechner law is illustrated in Fig. 3-2.
On a parallel but initially less conspicuous track runs the idea of a power
relationship between stimulus and response. Power relationships were
110 PSYCHOPHYSICS

TABLE 3-1 Representative values of Weber fractions

for different senses
Sense Weber Fraction, DI/I
Vision (brightness, white light) 1/60
Kinesthesis (lifted weights) 1/50
Pain (thermally aroused skin) 1/30
Audition (tone of moderate loudness) 1/10
Pressure (applied cutaneously) 1/7
Smell (odor of India rubber) 1/4
Taste (table salt) 1/3
Source: Geldard (1962).

Fig. 3-2 Relation between stimulus and sensation according to the Weber-Fechner law. The
logarithmic function illustrates the relationship between an arithmetic (sensation) and a geo-
metric (stimulus) scale.

originally proposed by the mathematicians G. Cramer and D. Bernoulli in their

consideration of the utility (subjective value) of money (Bernoulli, 1738). They
concluded that the subjective value of an increment of given size of money is
larger if it is added to a small amount than when added to a large amount, a
relationship that can be expressed with a power function. A power function
was proposed by Plateau to express his range partition results in which he had
several painter friends paint samples of the psychological midpoint between
a black and a white painted samples. From the remarkably uniform results and
Plateau’s assumption that there were ratios of stimuli and response involved,
he concluded that a power relationship could explain them. However, wanting
to establish a more detailed gray scale, he asked his friend Delboeuf to develop
 DIFFERENCES VERSUS MAGNITUDES 111

TABLE 3-2 Representative exponents of power functions relating subjective

magnitude to stimulus magnitude
Sensory Continuum Measured Exponent Stimulus Condition
Brightness 0.5 Point source
Brightness 0.5 Brief flash
Brightness 1.0 Point source briefly flashed
Lightness 1.2 Reflectance of gray papers
Visual length 1.0 Projected square
Redness (saturation) 1.7 Red-gray mixture
Taste 1.4 Salt
Smell 0.6 Heptane
Thermal pain 1.0 Radiant heat on skin
Heaviness 1.45 Lifted weights
Electric shock 3.5 Current through fingers
Source: Abbreviated from Stevens (1975). Reprinted by permission of the publisher.

more steps. The results were in better agreement with the Weber-Fechner
law, and the power law was largely forgotten until it found its champion in
Stevens.
Before Stevens it was J. P. Guilford who proposed in 1932 the first general
power law of psychophysics:

DI = ca I n , (3-5)

where I is the stimulus intensity, ca is the Weber fraction, and n is an exponent.

In this equation if n = 1, the relationship is logarithmic. Beginning in the 1930s
Stevens investigated the relationship between sensory and stimulus magni-
tudes of many kinds. He concluded that a power law connects all of them, with
the power variable anywhere from 0.25 (fourth root) to 1.5:

V = bx p , (3-6)

where V is the sensory magnitude, b is a constant, x is the intensity of the

stimulus, and p is an exponent. The implication is that the stimulus increments
proportional to equal sensory magnitude increments vary less than by equal
percentages as predicted by the Weber-Fechner law. Table 3-2 reproduces
selected power exponent values from a larger table by Stevens, and Fig. 3-3
graphically illustrates power functions with different exponents.
Stevens and other researchers found that subjects agree quite well on the
size of absolute magnitudes of sensory experiences as related to stimulus
magnitude. This can be done by direct judgment or magnitude production via
cross-modality matching. In the latter case the sensory magnitude is expressed
in another medium, for example, by comparing brightness of lights to loud-
ness of sounds or by squeezing a hand dynamometer to express the perceived
brightness magnitude.
112 PSYCHOPHYSICS

Fig. 3-3 Relation between stimulus and sensation based on three different power modulations.

From analysis of his own work and that of other researchers, Stevens con-
cluded that magnitude estimates of sensory experiences tended to have an
exponent near 1. In magnitude judgments the observer is free to assign any
chosen number to express the magnitude of the perceived stimulus. His own
results of magnitude judgments of the steps of a gray scale resulted in a power
exponent of 1.2 (Stevens and Galanter, 1957). When making category judg-
ments and difference judgments well above threshold, the power exponent
declined in size, and when making difference judgments near threshold, it
approached zero and the Weber-Fechner law.
Category scales have set numbers or adjectives and the results of category
judgments are usually nonlinearly related to those of magnitude judgments. A
gray scale based on equisection or category judgments, namely on differences,
results typically in an exponent of 0.25 to 0.5. In his own multiple bisection
experiment of brightness Stevens (1953) found an exponent of 0.26. Thresh-
olds are category measurements related to the uncertainty of the response,
with a relationship to the stimulus difference most accurately expressed as a
logarithmic function. The power function can also be closely matched with an
hyperbolic function, first implicitly used by Hering (1874):

ax
V= , (3-7)
1 + kx

where a and k are constants and x is the intensity of the stimulus. In a plot
with a linear ordinate and a log abscissa scale the resulting function has an S
shape. Thresholds, and the Weber-Fechner law approximating them represent
 PSYCHOPHYSICAL SCALING: LEVELS OF MEASUREMENT 113

TABLE 3-3 Effect of judgment type and size of difference on applicable stimulus
modulation
Type of Evaluation Applicable Power Example
Magnitude 0.8 and larger Lightness of gray papers
Paired comparison 0.05–0.75 Small color differences, Munsell chroma scales
Thresholds 0.01–0.33 Color thresholds

(in Fechnerian terms) local psychophysics; the power law is, in terms of
differences or magnitudes, representative of more global psychophysics.
Table 3-3 provides a comparison.
It is apparent that neither the logarithmic nor the power law provides an
explanation for the causes of the psychophysical relationships.They are merely
mathematical models. In the case of visual scaling the explanation for the
apparent signal compression is buried in the complexity of the visual system
and the possible impact of empiricism. While some details about the visual
system are known at this time, they are not nearly enough to develop a fully
detailed model predictive of all known effects. J. C. Baird has developed a
general simulation model that attempts to explain the effects described by the
laws from the aggregate action of neural cells with varying thresholds (sensory
aggregate model; Baird, 1997). It is not evident that this model applies to visual
scaling. K. Richter (1996) has proposed a model that attempts to explain visual
psychophysics only in terms of cone receptor saturation. It seems more likely
that post-receptor effects also play a role.
Psychophysics in the classical sense depends on so-called linking proposi-
tions.These are propositions of how neurophysiology and psychophysics might
be connected. Linking propositions can be strong or weak, depending on the
level of evidence supporting the proposition. A typical linking proposition is
that chromatic perception is supported by neurons with opponent response in
the visual area of the brain. Linking propositions related to the concept of a
uniform color space are still weak. The modern view of vision is that the brain
constructs what we consciously experience from the output of several of its
visual modules and using many (and probably many as yet unknown) decision
rules to construct the experience from generally ambiguous input at the retinal
level. As S. E. Palmer expressed it: “The job of visual perception is to combine
external and internal information to make meaningful facts about the envi-
ronment available to the organism” (Palmer, 1999). Simple links between
neural sensitivities to stimuli and psychological measurements or judgments
are therefore unlikely except in highly relativized situations.

3.4 PSYCHOPHYSICAL SCALING: LEVELS OF MEASUREMENT

An important aspect of psychophysics is the scaling methods used to assess

local and global scaling. Stevens (1946) proposed that onedimensional scales
114 PSYCHOPHYSICS

may be placed into a hierarchy with each subsequent type having greater
explanatory power than the previous one. This hierarchy is called levels of
measurement.

Nominal Scales
They are at the lowest level and refer to names or identifications of items only.
The same symbol is assigned to two things if they have the same value of the
attribute. An applicable sample for color is names. Colors can be grouped, for
example, into blues, browns, pinks, purples, and grays. An appropriate statistic
is the number of cases.

Ordinal Scales
Numbers are assigned to things in a way that the order of the numbers reflects
the order of the attribute. Items are placed into ascending or descending rank
order depending on some kind of magnitude. For colors a typical example is
a series of grays that can be ordered according to the concept of blackness.
The color nearest to black (sample E in Fig. 3-4) has ordinal scale value 1, that
closest to it but lighter (sample A) has value 2, and so on. A series of chro-
matic colors of varied chromaticness can also be placed into an ordinal chroma
scale. Ordinal scales do not contain any information about the sensory mag-
nitude of the steps between the grades on the scale. In Fig. 3-4 a random

Fig. 3-4 Twelve random samples of gray papers (A–K) are placed into an ordinal scale based
on their luminous reflectance value Y and on an interval scale based on lightness L*.
 PSYCHOPHYSICAL SCALING: LEVELS OF MEASUREMENT 115

collection of gray samples has been ordered according to the luminous

reflectance scale (Y). Luminous reflectance places the samples into an ordinal
scale where the scale values do not provide any information about the per-
ceptual magnitude of the steps between the grades of the scale. An ordinal
scale is subject to logical operations: equal to, greater than, less than; and
appropriate statistics are median, percentiles, or rank-order correlation.

Interval Scales
At the next level interval scales provide quantitative information concerning
the distances or differences between grades. Here things are assigned numbers
so that differences between numbers reflect differences of the attribute. In an
interval scale two grades differing by three interval units at the lower end of
the scale, and two grades at the higher end of the scale also differing by three
interval units are equally distant (e.g., the Celsius and Fahrenheit temperature
scales). In interval scales the distance between two percepts is represented
with a number according to

a = mb + c0 , (3-8)

where a is the scale number representing a grade, m is any positive number,

b is the distance of the grade from the neighboring grade or the origin, and c0
is any finite number. Addition of and multiplication with constants are per-
missible transformations and statistics include mean, standard deviation, or
correlation coefficient. Color scales usually are interval scales. In Fig. 3-4 the
twelve random gray samples have also been ordered according to the interval
scale L*. The perceptual difference between samples B and I is about half as
large as that between samples F and D.

Ratio Scales
These are interval scales that have a natural origin. In this case things are
assigned numbers such that differences and ratios between the numbers reflect
differences and ratios of the attribute. While in interval scales two numbers
could be applied arbitrarily to the scale (m and c0) only one number can be
arbitrarily assigned:

a = mb. (3-9)

Examples are lengths in meters,duration in seconds,and temperature in degrees

Kelvin. As mentioned before, ratio scales in regard to color are controversial.
Ratio scales can be multiplied with a constant only and statistics include percent
variability.
116 PSYCHOPHYSICS

Absolute Scales
Here things are assigned numbers such that all properties of the numbers
reflect analogous properties of the attribute. These measurement levels are
part of a continuum of order, with the level of order being weakest for nominal
scales and strongest for absolute scales. In practice, depending on the mea-
surement technique used, the same scale can have properties of two mea-
surement levels.

3.5 SCALING METHODS

Many different methods of scaling have been developed in psychophysics. For

absolute or difference thresholds the classical methods discussed in this
section are those described by Fechner (1860).

Method of Adjustment or Method of Average Error

In this method the observer can adjust the stimulus magnitude and does so
until she observes a just noticeable difference. On the surface this is a simple
and straightforward method. However, it requires equipment on which the
stimulus magnitude (under control of the observer) can be continuously
adjusted. The results are direct but also subject to cognitive adjustments and
biases.

Method of Limits or Method of Minimal Change

Here the experimenter presents changes in the stimulus magnitude in preset
small increments in ascending or descending order, starting with an imper-
ceptible increment in the former case and a clearly perceptible one in the
latter. This continues until the observer indicates the perception of a differ-
ence in the former case or the absence of a difference in the latter. The JND
is taken to be represented by the average of ascending and descending trials.

Method of Constant Stimuli

In this method the experimenter selects several constant pairs of stimuli
ranging from below threshold to above threshold. These are then presented
several times randomly to the observer. The observer responds with a yes or
no, and the JND limit is determined at the 50% or perhaps at the standard
deviation level. Alternatively, the experiment is set up so that the observer
must respond to a forced choice between two pairs of stimuli. This method has
been further varied in the so-called staircase procedure. Here a forced choice
is imposed at the lowest level. If the observer responds wrongly, the stimulus
difference is increased for another forced choice. The JND limit is approached
from both sides in such experiments.
 UNIDIMENSIONAL SCALING METHODS 117

Matching Method
A well-known experiment, believed related to color thresholds, is the deter-
mination of color matching error by MacAdam (1942). He constructed an
apparatus in which color stimuli could be varied along selected lines in the
CIE chromaticity diagram at constant luminosity. A standard stimulus was dis-
played and the observer adjusted a test field until the two fields matched, that
is, had identical appearance. From the variability of repeated tests MacAdam
calculated the matching error that he thought to be related to the difference
threshold.
These methods are known under the general rubric of confusability scaling.
Thurstone stated a law of comparative judgment in 1927. This law considers
every perceptual magnitude judgment as a variable datum from a discrim-
inable process that he took to be normally distributed. In this manner the full
power of statistics of normal distribution can be applied. Under specific con-
ditions such statistical treatment results in confirmation of the Weber-Fechner
law, other conditions result in applicability of Steven’s power law.

3.6 UNIDIMENSIONAL SCALING METHODS

Unidimensional scaling involves perceptions that have one attribute only.

Several physical dimensions may be involved in generating the attribute. Over
the century and a half since the beginning of psychophysics several scaling
methods for unidimensional scaling of perceptual distances have been devel-
oped. Only those used in color scaling will be briefly discussed.

Partition Scaling
First used in the form of the equisection method by Plateau, partition scaling
is a direct estimation method. In the equisection method the perceptual dis-
tance between two different stimuli may be halved in several steps. Another
version determines equal-appearing intervals.

Ratio Production and Ratio Estimation

As the name implies, the method results in a ratio scale. In ratio production
the observer adjusts a magnitude of a perceptual attribute until it equals the
perceived double magnitude (or other ratio) of a reference. In ratio estima-
tion the observer estimates the ratio between reference and test stimuli. For
reasons discussed previously, this method is rarely applied in color scaling.

Magnitude Production and Estimation

In production the observer is given a stimulus and a number that corresponds
to it and is asked to produce a stimulus that is representative of another
118 PSYCHOPHYSICS

number. Thus a certain sound may be said to represent the value 10, and the
observer is asked to modify the sound stimulus so that the resulting loudness
of the sound represents the value 15, say. In magnitude estimation the observer
estimates the perceived magnitude of experiences from stimuli and assigns
numbers to them. For color scaling a large number of stimuli or the capabil-
ity of stepwise or continuous stimulus adjustment is required.

Category Scaling
Here the observer is asked to separate large numbers of experiences into
categories. The corresponding samples must be similar enough so different
observers arrive at different categorization. The variability in judgment by dif-
ferent observers is assumed to follow a standard normal distribution from
which an interval scale can be constructed. A typical example is acceptability
or pass-fail judgments of small color differences in which the observer is asked
to determine if a given sample meets or fails a criterion of acceptability.

Paired Comparison
In this method all samples are presented to the observer in all possible pairs
or in all pairs of test against a reference sample. The proportion of times a
given sample is judged greater in magnitude of a given attribute is determined.
Interval scales are derived from the results under the assumption of statisti-
cally normal distribution.
In recent years a form of interval judgment for the purpose of suprathresh-
old color scaling has been in wide use. In this method sample pairs exhibiting
small differences are compared against a reference difference, usually in form
of an achromatic pair with a perceptual difference of similar magnitude to
those of the sample pairs under estimation. Alternately, an International Stan-
dards Organization (ISO) type gray scale that displays achromatic pairs with
varying perceptual differences has been used. In such situations what are mul-
tidimensional color differences (usually involving hue, chroma, and lightness
differences at the same time) are evaluated as if they were unidimensional.
The surround lightness and chromaticness affects the perceptual magnitude of
the reference pair(s), as well as of the test pairs making the result a function
of the surround.

Minimally Distinct Border Scaling

A novel method of scaling color differences was developed in the 1960s and
1970s by R. M. Boynton and P. K. Kaiser (Boynton, 1983). Their idea was that
with two perfectly juxtaposed color fields at equal luminance the strength of
the resulting border should be an indicator of the chromatic difference in the
constant luminance plane. The method was found to be applicable only where
the two fields differed in activation of the L and M cones.
 MULTIDIMENSIONAL SCALING 119

Fig. 3-5 Example of a psychometric function. Samples contain different amounts of a red and
a gray stimulus (by spinning disk mixture). Observers judged if a sample contained more red
than the one shown previously. Circles indicate results when a reference sample containing
80% red was shown immediately before the test sample, triangles where no reference was
shown. From Stevens (1975).

3.7 PSYCHOMETRIC FUNCTION

Psychometric functions are the functions expressing the relationship between

subjective and objective measures. The usually normal statistical distribution
of individual judgments generally results in nonlinear psychometric functions.
An example is pass-fail judgments of color difference. Here the relationship
between percent acceptability (% pass) and calculated color difference was
found to have a typical S-shaped (sigmoidal) form (see Fig. 3-5 for another
example). In such cases it is necessary to linearize the visual scale by using an
appropriate method (e.g., Indow and Morrison, 1991).A function that has been
used in recent years for this purpose is the cumulative normal distribution
(probit) function (e.g., see Berns et al., 1991).

3.8 MULTIDIMENSIONAL SCALING

Multidimensional scaling (MDS), originally developed by W. S. Torgerson in

1952 (1958) and R. N. Shepard (1962), creates geometric models based on sim-
ilarities, dissimilarities, or proximity. Mathematical analysis is performed on
spatial or distance data. However, MDS can also be applied to nondistance
data. It can address the nature of the metric of the multidimensional stimulus
space implicit in the data, meaning it can extract the unknown dimensional-
ity. Because psychological data usually contain considerable variability differ-
120 PSYCHOPHYSICS

Fig. 3-6 Configuration of MDS analysis of judged differences between fourteen spectral colors
(circles with numbers and identified by spectral wavelength in nanometers. A circle segment
(closed to a full circle) has been fitted to the points. Presumably on the segment between spec-
tral colors 1 and 14 extraspectral red and purple colors mixed from various ratios of 1 and 14
would be located. From Shepard (1964).

ent MDS techniques often find several different models, all with comparable
accuracy of fit. A widely used methodology of MDS is INDSCAL (individual
difference scaling; Carroll and Chang, 1970). It is based on euclidean geome-
try, but evidence has been found that it can “adequately” recover sample con-
figurations even if the true metric is non-euclidean.
In one version of MDS the pairwise distance (in terms of a visual estimate
or any other distance measurement) between all items used in the test is estab-
lished. From these data a similarity or dissimilarity matrix is created as input
to the MDS module. Several parameters or assumptions can be changed in
MDS and the results are subject to interpretation.
An interesting finding of MDS is that when perceptual distances between
fourteen spectral colors are judged by observers the result of the MDS analy-
sis of the data is a two-dimensional structure that is fit best by a circle segment
(Fig. 3-6; Shepard, 1994).
MDS has been applied extensively to Munsell data by T. Indow and co-
workers (Indow, 1988) and others. Using judgments of magnitude of color dif-
ference between many samples of the Munsell system as input into the MDS
analysis, Indow recovered the psychological diagram of the Munsell system
with good accuracy (see Fig. 4-4 in the next chapter). In principle, scaling of
 PSYCHOLOGICAL AND PSYCHOPHYSICAL SPACES 121

small complex color differences (involving more than one attribute) might be
analyzed by MDS to determine the implicit dimensions. This does not appear
to have been done so far, perhaps because the dimensionality of the color
experience is believed known.
The MDS method is not without its critics (see e.g., Saunders and van
Brakel, 2002). Some criticism harks back to Kries original doubts about psy-
chophysics, namely are magnitudes and differences distances that can mean-
ingfully be expressed in numerical or geometrical form.

3.9 PSYCHOLOGICAL AND PSYCHOPHYSICAL SPACES

Complex psychological phenomena such as color vision are usually illustrated

in geometrical spaces. Such spaces presuppose an isomorphism between color
experiences and the selected geometrical model. The isomorphism may be
based on the logic of the four unique hues and their mixtures as well as of
hues, black and white or on equality of differences. In most color order
systems, as seen in Chapter 2, white is placed on top producing the isomorphic
association lighter equals higher. Greener means closer to the location of
unique green, and so on. Closer to the centerline means grayer. The specific
isomorphism depends on the distance criterion used. In a perceptually uniform
(for a specific set of conditions) space equal distances between points repre-
sent equal perceived differences.There are two kinds of isomorphism in regard
to color space: (1) between the perceptual experiences and the psychological
color space and (2) between the psychological and the psychophysical space.
As seen in Chapter 2, historically color solids had simple geometrical forms
with euclidean geometry (cones, spheres, pyramids, cubes).The Munsell system
indicated that for a “uniform” system the underlying structure might be simple
(cylindrical) but the surface of the solid complex. As will be seen in Chapters
7 and 8, a euclidean form of a uniform color space can be ruled out, at least
at the global level.
The concept of space is of unlimited complexity. While most humans cannot
imagine a space with more than three dimensions mathematicians have devel-
oped a number of categories of spaces. C. F. Gauss and N. I. Lobachevski
invented hyperbolic geometry and G. F. B. Riemann the elliptic geometry
named after him. Riemannian geometry has been used in the past in connec-
tion with color scaling (e.g., MacAdam, 1981; Völz, 1998b). In the hyperbolic
geometry the sum of the angles in a triangle is less than 180° degrees, in the
elliptic geometry greater than 180° degrees. In string theory modern physicists
have proposed a universe of nine dimensions.2
As will be shown, a uniform color space based on small color differences
requires an elliptic geometry. Psychophysical spaces that have no claim on uni-
formity, such as the Rösch-MacAdam space, fit into euclidean geometry
without difficulty. It is evident that any kind of three-dimensional euclidean
space form and associated color solid can be used as a regular (but not
122 PSYCHOPHYSICS

uniform) psychophysical color space. Modern color difference formulas

dealing with some of the complexities of uniform color space currently are
based on mathematical modifications of a euclidean space. The exact type of
geometry applicable to a uniform color space under well-defined conditions
of observation and magnitude of differences remains to be determined.

3.10 PSYCHOPHYSICAL SCALING AS A BASIS OF COLOR SPACE

Color space can be represented in a continuous fashion only with mathemat-

ical formulas that, for most people, lack direct comprehension in terms of color
experiences. Historically color solids have been illustrated with two- or three-
dimensional arrangements of color samples in form of atlases or “color trees.”
Here psychological and psychophysical scaling is essential if psychological uni-
formity relative to specific conditions is the goal. Psychophysical scaling has
the advantage of providing physical measurement support for the stimuli. On
the other hand, they require a psychophysical model that accurately reflects
psychological results. A major impetus for psychophysical scaling of threshold
and small suprathreshold differences and related mathematical models has
been the desire for objective quality control of colored materials.
Psychophysics as a methodology is essential for the construction of a
uniform color space. Psychophysical data are subject to considerable variation
based on the physics of measurement and observer variability as well as vari-
ability in observational context. In the case of color attribute and color dif-
ference scaling, it is not yet clear what the major contributors to variability
are. Careful experimentation is required to determine the effects of individ-
ual contributors.
 Chapter 4
Color Attributes
and Perceptual
Attribute Scaling
4.1 THEORIES OF VISION

Vision is defined as the sense, mediated by the eyes, by which the positions,
qualities, and movements of objects are perceived. The generally accepted
view is that patterns of light energy are absorbed by light sensitive cells in the
eye. The resulting electrochemical signals are passed into the brain where they
cause activation of other cell types eventually resulting in sensations and per-
ceptions of form, color, and movement. A representation of the outside world
is generated in the retina and eventually in the brain. There are standard rela-
tionships between stimulus and resulting perception for which, presumably,
neural correlates exist in the brain. Exceptions to these standard relationships
are just that. They are caused by limitations in the neural apparatus. Much of
the efforts of visual science in the 1980s and 1990s have been attempts at elu-
cidation of these standard relationships and exceptions and finding their
neural correlates (e.g., see De Valois, 2000). But parts of this view face serious
problems from a slew of perceptual responses that are not explained by
standard models.
Already Helmholtz thought that perception is the result of the interaction
of the neural messages from the eyes with stored memories from past experi-
ences (Helmholtz, 1866). With the wider acceptance of Darwin’s ideas about
evolution in the twentieth century, some researchers began to look at senses

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

123
124 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

as tools of animals to cope with challenges in their particular ecological niches.

In the second half of the twentieth century the American psychologist J. J.
Gibson developed what he called an ecological approach to visual perception.
He believed that the important rules of vision are found in natural viewing
environments and not in fixed laboratory situations. Indeed, both are realities
and ultimately must find meshing explanations. But it appears quite clearly
wrong to assume that there is a simple standard relationship between states
of retinal and early neural cells of the visual system and percepts. There is no
doubt that vision has provided its bearers with an important tool to help make
decisions affecting their own life and that of their offspring. Vision, imple-
mented in many different ways in the animal kingdom, is undoubtedly shaped
by evolutionary forces. It remains to be determined how.

4.2 HISTORICAL DEVELOPMENT OF VIEWS ON ATTRIBUTES

Regardless of how a particular color experience is generated, it can be said to

have certain attributes, that is, inherent characteristics. The idea of color attrib-
utes has developed slowly over an extended period of time. As shown in
Chapter 2, ancient Greeks appear not to have had a clear concept of chro-
matic color attributes. Only the concepts of light and dark and of hue were
familiar to them. The simple linear color scales of the philosophers, originally
seemingly in random order, later implied an ordering by lightness, without
regard to hue ordering in the sense of the spectrum. In Aristotle’s seven-color
scale white precedes yellow followed by crimson, violet, green, and blue which
precedes black. The red-centered five-color scale, found in descriptions until
the sixteenth century, is that of Chalcidius: white–yellow–red–blue–black.
From observations in nature and the work of artists the lightening and dark-
ening of full colors (or of pure highly chromatic pigments) was well known in
artist’s circles of the twelfth century, as the text by Theophilus demonstrates.
A painting technique of the early Italian Renaissance, described by Cennini
in 1390, involved the use of pure chromatic pigments for the darkest areas and
dilutions with a white pigment for lighter areas. In 1435 Alberti advised his
fellow painters as follows: “. . . you may change the color with a little white
applied as sparingly as possible in the appropriate place within the outlines of
the surface, and likewise add some black in the place opposite to it. With such
balancing, as one might say, of black and white, a surface rising in relief
becomes still more evident. Go on making similar sparing additions until you
feel you have arrived at what is required.” The full complement of tonal colors
began to be exploited by some artists wanting to avoid a certain garishness
that can arise when painting with pure chromatic pigments. Chief among them
was Leonardo da Vinci who developed the sfumato (smoky) style of paint-
ing and who is credited by some art historians as having distinguished be-
tween lightness (chiarezza in Leonardo’s terminology) and chroma (bellezza)
(Ackerman, 1980). Quantitative estimates of the natural lightness of full object
colors were provided in print first by Cardanus in 1563, as seen in Chapter 2.
 HISTORICAL DEVELOPMENT OF VIEWS ON ATTRIBUTES 125

As has been discussed earlier, it is reasonably certain that Forsius and Glisson
did not have a concept of a third attribute, beyond hue and lightness.
The concept of saturation as an attribute of prismatic color makes its defin-
itive appearance with Newton. It is clearly based on his experiments with over-
lays of various wavebands of light. He described points on a radial line in his
diagram (see Fig. 2-10) as indicating colors “proportional to the fullness or
intenseness of the [prismatic] Colour, that is, to its distance from whiteness.”
Newton recognized seven primary and an indefinite number of intermediate
hues. He described lightness in connection with colorants as follows: “Now
considering that these grey and dun Colours may also be produced by mixing
whites and blacks, and by consequence differ from perfect whites not in
Species of Colours but only in degree of luminousness, it is manifest that there
is nothing more requisite to make them perfectly white than to increase their
Light sufficiently . . .” (Newton, 1704, p. 112). Here we see for the first time (if
not in the same place in the book) mention of three attributes describing color
perceptions.
Mayer, who in the mideighteenth century developed the first plan for a
three-dimensional color order system, did not distinguish color series from the
most saturated colors on the surface of his double triangular pyramid through
the interior of the triangle. A possible reason is that in his own efforts with
pigments he ended up with a neutral gray quite removed from the gravimet-
ric center of the triangle, according to his own information at r3 g2 b7, indicat-
ing that his scheme of determining relative strength of pigments was less than
perfect.
Also Lambert did not recognize saturation or chroma as a separate color
attribute, even though it can be seen as implicit in his single pyramid. The
reason is similar to the one applicable in the case of Mayer. His three primary
colors are placed on the same level regardless of their lightness, and his choice
of colorants and mixture ratios did not result in black falling onto the gravi-
metric center of his basis triangle but, again, not far from the blue corner.
The painter Runge provided clear and explicit discussion of the general idea
of saturation, but without using such a term. Because he placed the full colors
on the circumference of the central horizontal plane of his color sphere, its
vertical dimension does not refer to brightness or lightness. But Runge under-
stood clearly the desaturation of a full color, by the admixture of appropri-
ately selected complementary colors or combinations: “When we add to pure
green, a product of yellow and blue, the smallest amount of red as the third
color, we learn that it simply destroys and dirties the pleasant appearance of
green without adding the appearance of redness. Therefore, through a stronger
admixture of red, green is dissolved into completely colorless smut, or gray,
and assumes a reddish hue only through an even stronger admixture. . . . All
diametrically opposed colors and mixtures on the circle are (in the center point
of the circle) dissolved (into gray)” (Runge, 1810). In the colored figure of
the equatorial cross section of his sphere (see Fig. 2-22) Runge shows four
desaturation steps toward the central gray for each full color.1
Grassmann, as we have seen in Chapter 2, supplied Newton’s diagram with
126 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

a mathematical foundation based on Helmholtz’s and Maxwell’s ideas of three

fundamental color processes. He also, for the first time, explicitly referred to
three attributes that combine to form a color perception: “If, finally, we con-
sider a light of arbitrary composition, the eye can distinguish in it only the
three mentioned attributes, that is, any impression by a light can be imitated
by mixing a homogeneous color of a certain intensity with white light of a
certain intensity. Therefore, we have to distinguish three things in each impres-
sion: intensity of color, hue, and intensity of the admixed colorless light”
(Grassmann, 1853, pp. 70–71). Grassmann provided a general connection
between perceptual color attributes and physical magnitudes.2
In his Physiologische Optik, preceding the Handbook, Helmholtz slightly
modified Grassmann’s definition:3

Every impression on the eye made by an arbitrarily mixed light can always be
represented as a function of three variables that can be expressed in numbers,
that is,
1. Quantity of saturated colored light,
2. Quantity of white light that when admixed results in the same color
impression,
3. Wavelength of the colored light. (Helmholtz 1860)

In the Handbook he described the three attributes in the form still in use:

Accordingly, with all possible combinations of systems of aether-waves of dif-

ferent frequencies of vibration, there is after all a comparatively small number
of different states of stimulation of the organ of vision which can be recognized
as different colour sensations. First of these are the series of saturated colours,
composed of the colours on the spectrum, along with purple which links the ends
of the series. Each of these hues again may occur more or less pale in different
gradations. The paler it is the less saturated it appears. . . . Thus, we have here
two kinds of differences between colours, namely, first, differences of hue, and
second, differences of saturation. . . . Lastly, in ordinary speech we are wont to
describe differences of luminosity as differences of colour, however, only in case
colour is considered as a characteristic of bodies. (italics in the original)

In 1866 Brücke presented a schematic color sphere and a cross-sectional

view in which he showed presumed lines of constant saturation (Fig. 4-1). He
viewed constant saturation contours as representing the surfaces of spindle
shapes, with varying ratios of length to width, ending in the sphere shape at
maximum saturation for a given level of lightness.
In Wundt’s color sphere of 1874 hue (Farbton), saturation (Sättigung), and
brightness (Lichtintensität) are established concepts. His student Kirschmann
published in 1895 a paper “Color-saturation and its quantitative relations” in
which he described the design of a color disk that, when spun with sufficient
speed, displayed what was described as fifteen perceptually equally different
saturation steps of the test color of the disk (Fig. 4-2). The heart-shaped figure
A is colored with the test color (e.g., an approximation of a Hering type full
 HISTORICAL DEVELOPMENT OF VIEWS ON ATTRIBUTES 127

Fig. 4-1 Brücke’s vertical section through the color sphere with white on top and black on the
bottom, 1866. The full interior lines are meant to be lines of constant saturation.

color). The shapes of B and C are designed to produce from black and white
the appropriate level of gray to be added to the full color so that all resulting
saturation steps have identical lightness (thus representing a chroma scale
when seen as object colors). The exact shape of B and C depends on the pho-
tometrically measured luminosity of the color of A. With this method
Kirschmann investigated the validity of the Weber-Fechner law in regard to
saturation.
In Hering’s system full colors occupy one corner of equilateral triangles.
The remaining colors in a triangle consist of those of the gray scale and the
veiled (verhüllte) colors with the same hue as the full color. As mentioned in
Chapter 2, Hering was fully aware of the varying lightness of full colors but
never explicitly described how lightness as an attribute was to fit into his
“natural color system.” Any color could be described as the sum of one or two
primary colors, white and black. Thus, in the majority of cases, colors had four
attributes, for example, for a brown yellowness, redness, whiteness, and black-
ness. Hering abandoned use of the term saturation because he believed its
meaning to be “contaminated” by Helmholtz. He defined colors of “equally
strong veiling” as those falling on lines parallel to the line connecting white
and black in his triangle,4 those having the same ratio of full color to total color
128 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

Fig. 4-2 Construction of a disk arrangement for the spinning disk mixture resulting in fifteen
concentric circles at equal luminance with perceptually equally (according to the Weber-Fechner
law) different steps of saturation of the color displayed on shape A. From Kirschmann
(1895).

content. He commented on the difficulty of comparing degree of veiling for

differently hued colors and does not seem to have regarded it as a primary
attribute of color perception.
Later the Hering pupil E. G. Müller and the psychologist D. Katz suggested
replacement of the term saturation with Eindringlichkeit (penetrance). In 1917
the psychologist C. Stumpf completely rejected saturation as an attribute and
believed it to be a cognitive abstraction for the resemblance of a color per-
ception to its ideal.
Munsell, in his effort to develop a useful teaching tool, first relied on the
form of a new kind of color sphere, developed by him in 1898. This sphere dif-
fered from Runge’s in that the vertical dimension has an unambiguous defin-
ition as lightness (value) with the result that on this sphere Hering’s full colors
are not located on the equatorial plane. Munsell had studied art in Paris
around 1880 and, perhaps influenced by the French use of valeur in tonal paint-
ing, settled on the term value for lightness. The Greek term chroma (surface
skin, color) was also in use as a general term for color in French art literature
at the time of his studies (e.g., George Seurat’s chromo-luminarisme painting
style5) and it appeared in the title of Rood’s influential Modern chromatics. It
seems intuitive to use it in a quantitative sense to indicate how much chro-
matic color is present. Munsell’s three color attributes, principally in agree-
 WHITENESS AND BLACKNESS 129

ment with those of Helmholtz, were thus named hue, value, and chroma.
Munsell, intrigued with the decimal system, selected a 100-part hue circle and
five primary colors rather than Hering’s four (he thought Hering to be wrong
because Hering’s claims for the operation of his opponent-color system were
not in accord with contemporary knowledge about the retinal structure). His
value scale consisted of ten steps. The chroma scale was open ended because
Munsell soon found that he could not fit all colors with uniform chroma
spacing into a sphere form. As described below and in Chapter 7, scales for
all three attributes were intensively investigated before finding their final form
in the Munsell Renotations.
Ostwald, working in isolation during the First World War, erroneously
believed Helmholtz had made no distinction between what we now call bright-
ness and lightness.6 For this reason he rejected lightness as an attribute for his
system and followed Hering in using full color content, whiteness and black-
ness. But unlike Hering, he based his hue circle on three primary colors, yellow,
red, and blue. Object colors were defined as the sum of full color perceptions
and perceptions of white and black. Hering’s colors of equally strong veiling
were termed by Ostwald Reingleiche (colors of equal purity). Analysis of
Ostwald’s system in 1944 indicated that:

1. The system did not cover the full range of object colors as defined by
the Rösch-MacAdam limits.
2. Ostwald’s claims for complementarity and visual uniformity in terms of
hue differences around the hue circle have not been met and cannot
be met.
3. Various compromises have been made by Ostwald in physically imple-
menting his system (Foss, Nickerson, and Granville, 1944).

Basic ideas of Hering were also used by Johannsson and Hesselgren and
resulted in the development of the Swedish Natural Color System (see Chap-
ters 2 and 7).

4.3 WHITENESS AND BLACKNESS

In Chapter 2 it was mentioned that white and black have been considered
colors for some 3000 years or more. But there has been a certain amount of
controversy about this subject. Non-hued visual percepts are often called
achromatic colors, that is, colorless colors. Truly achromatic colors differ from
chromatic colors in the lack of the key perceptual attribute of hue. While hued
colors are well defined, unique blue is a bluish color that is neither greenish
nor reddish, neither black nor white are well defined in a similar sense. This
becomes apparent if one looks, for example, at a collection of white and
black textiles. Generally preferred white and black both have a slightly bluish
130 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

tinge. But the preferences differ among individuals and cultures. In addition,
while not any chromatic stimulus can appear unique red, any neutrally reflect-
ing surface can be made to look white or black or any gray in between depend-
ing on illumination and surround. Perceptions of white and black as the
achromatic extremes are outputs of our color vision system, just as percep-
tions of hue and chroma are. It appears entirely justified to consider achro-
matic colors to be on an equivalent level with chromatic colors.
Helmholtz defined black as the absence of light energy, while Hering,
correctly, defined it as an indirect manifestation of light in terms of contrast.
When looking at reflectance data of pigmented black and white paint layers,
it is obvious that while the white layer reflects most light the black layer
reflects very little. But the small amount it reflects is responsible for the fact
that a black pigment layer, selectively illuminated with high intensity white
light, can be made to look white (as demonstrated by A. Gelb in 1929). Con-
trast plays a key role in the perception of black, as is well known from the
demonstrations of the effect of surround brightness by Evans and from
the fact that an unpowered television screen is gray and not black in appear-
ance. The blackness we see in images on the screen must come from contrast
effects.
The perception of blackness has been studied in considerable detail (e.g.,
see Volbrecht and Kliegl 1998). The blackness induction curve is somewhat
similar to the heterochromatic flicker brightness curve (see Chapter 5).
However, it varies significantly depending on the design of the experiment. If
a monochromatic ring surrounds a white center, the agreement with the flicker
brightness curve is relatively high. If center and ring are reversed, the func-
tion appears to have a chromatic opponent color component added, reminis-
cent of the Helmholtz-Kohlrausch effect. A detailed model of blackness
perception based on cone sensitivities and assumptions about the physiology
of color vision has been developed by K. Shinomori and co-workers
(Shinomori, 1997).
As we have seen, whiteness as a psychological attribute has been defined
in terms of a perceptually uniform gray scale by Hering and Ostwald. The psy-
chophysical description of whiteness has been of interest since the mid-1930s.
The definition of the portion of color space that most observers call white or
near white, and its description in terms of a psychophysical formula, have been
driven by industrial interest in measuring whiteness of papers or textiles.
Several formulas were developed around 1960, and in 1979 the CIE recom-
mended a whiteness formula to be used in the interest of uniformity (CIE
1979; for a bibliography until 1976, see Sève, 1979). An improved formula is
under development. It is of interest to note that the preferred psychological
optimal white is not one representing perfect 100% reflectance across the
spectrum, nor an equal energy distribution, but one with a slight bluish tint
and achieved with the addition of fluorescent whitening agents. This indicates
that whiteness and lightness are different psychological and psychophysical
concepts.
 EVANS’S FIVE COLOR ATTRIBUTES 131

4.4 EVANS’S FIVE COLOR ATTRIBUTES

In 1974 R. M. Evans published his interpretation of the results of many years

of study of the appearance of color. He concluded that there are five attrib-
utes that require consideration for a full description of color appearance.These
are brightness, brilliance, lightness, saturation, and hue. Brightness is conven-
tionally defined in terms of luminosity. Brilliance is a term that encompasses
the scale of perception from grayish colors, through the zero grayness (G0)
point to fluorent colors (a term proposed by Evans). Object colors are seen to
contain grayness up to a certain level of luminous reflectance (dependent on
the surround luminous reflectance) where the zero grayness point is located.
At this point they are seen as equally light as the surround. When the lumi-
nous reflectance increases further, the colors are seen as lights (weaker or
stronger, depending on the degree of fluorence). The spectral zero grayness
function (Fig. 4-3) bears close resemblance to the saturation discrimination
function, a function describing, based on dominant wavelength, the purity of
Munsell samples at constant chroma, as well as to MacAdam’s color moment
per lumen function (MacAdam, 1981). Evans defined lightness as relative
luminance and saturation in a form comparable to colorimetric purity (see
Glossary). In his system saturation is the perception of the brilliance of the
hue component in relation to the total brilliance of the color. Evans defined
Munsell chroma as brilliance difference from a gray of the same value. He

Fig. 4-3 Zero grayness (G0) of object colors as a function of dominant wavelength. From Evans
(1974).
132 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

described the problems with Judd’s original uniform color space definition as
due to the need for four dimensions of such a space. When defining uniform
color space in terms of surrounds that are always intermediate to the two
colors compared, as Judd did in his re-definition, the four-dimensional space
can be reduced to three dimensions. This minimizes the required stimulus
increment or decrement for color differences. Evans believed the spacing of
the Munsell system to be a good approximation of Judd’s re-defined uniform
color space. The usual approach, however, is to base an attempted uniform
color space on a single achromatic surround of a given lightness.

4.5 COMMON COLOR ATTRIBUTE DEFINITIONS

The three psychological Munsell attributes have become generally accepted

for object colors. Wyszecki termed the empirical evidence of three attributes
being sufficient for an observer with normal color vision to describe any per-
ceived color “overwhelming.” His terms for the three attributes were hue,
lightness, and chromaticness (Wyszecki, 1981).

Hue
The term hue is defined as “attribute of visual perception according to which
an area appears to be similar to one of the colors red, yellow, green, and blue,
or to a combination of adjacent pairs of colors considered in a closed ring”
(CIE, 1987). Essentially this definition indicates that hues represent the vari-
able one experiences when looking at a Munsell (or other system) hue circle.
As mentioned earlier, such hue circles derive their legitimacy from the
arrangement of hues in the spectrum, as well as systematic mixtures of stimuli
from the beginning and end of the visible spectrum (see Fig. 3-6).

Brightness, Lightness
Lightness is an attribute related to brightness, the definition of which is:
“attribute of a visual perception according to which an area appears to emit,
or reflect, more or less light” (CIE, 1987). Brightness is generally taken to
apply to light sources. Lightness is defined as: “the brightness of an area judged
relative to the brightness of a similar area that appears to be white.” Lightness
thus can be said to be relative brightness and is generally taken to apply to
object colors. Brightness and lightness have been controversial since their def-
initions because they have been defined based on additive functions using an
experimental method that is far from natural. As will be shown in Chapter 5,
brightness and lightness of chromatic lights, respectively objects, when viewed
in natural or strongly relativized conditions have additional components, not
contained in the additive function.
 COMMON COLOR ATTRIBUTE DEFINITIONS 133

Saturation, Chroma
Over the years there has been considerable discussion concerning the term
used for the third attribute. Helmholtz’s Sättigung, or saturation, is now an
“attribute of a visual sensation that permits a judgment to be made of the pro-
portion of pure chromatic color in the total sensation” (CIE, 1987). The term
chroma is closely related to Munsell chroma and refers to an “attribute of
visual sensation that permits a judgment to be made of the amount of pure
chromatic color present, regardless of the amount of achromatic color.” A
differently worded definition is: “attribute of color used to indicate the
degree of departure of the color from the gray of the same lightness” (ASTM
E284). This definition relates directly to the Munsell system. As discussed
next, chroma can be seen as related to an attribute termed “colorfulness,” as
lightness is related to brightness. Chroma and saturation are identical for two
colors having the same hue and lightness. Saturation remains constant regard-
less of brightness or lightness. Chroma, on the other hand, increases as
lightness increases. Saturation relates to an inverted cone arrangement of
colors, for example, the DIN system, while chroma refers to a cylindrical
arrangement.
The question as to how intuitive the concept of chroma is to the average
observer is an interesting one. We have seen that as a well-defined concept, as
introduced by Munsell, it is only approximately 100 years old. Seemingly there
are no terms of folk psychology that directly refer to it. Empirical evidence of
the author with more than 100 untrained observers in experiments where they
had to sort Munsell chips of constant hue into a Munsell value/chroma tem-
plate indicates that many had considerable difficulty to do so because of uncer-
tainty about the concepts of lightness and chroma. A recent investigation has
confirmed the higher uncertainty of judgments of assessing color differences
as chroma differences, compared to lightness and hue differences (Melgosa
et al., 2000).

Hunt’s 1977 Proposals

In 1977 R. W. G. Hunt proposed new systematic terminology for color attrib-
utes at four levels:

1. Perceptual (psychological)
2. Psychophysical (related to stimulus)
3. Psychometric (interval scales)
4. Psychoquantitative (ratio scales)

Perceptual terms include brightness, lightness, hue, saturation, and perceived

chroma. The psychometric terms have the modifier “metric” before them. It
changes to “quantitative” for the psychoquantitative terms. The psychoquan-
titative terms are controversial because of the philosophical problems sur-
134 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

rounding them. Hunt also proposed revised definitions for chroma and a new
attribute: colorfulness. It is defined as the “attribute of a visual sensation
according to which an area appears to exhibit more or less chromatic color.”
Colorfulness refers to chromatic power perceived regardless of the magnitude
of the stimulus. Hunt has used, for example, theatrical lights of varying inten-
sity to demonstrate the meaning of colorfulness. Hunt proposed the definition
of perceived chroma as amount of chromatic color in related colors judged in
proportion to the average brightness of the surroundings. His proposals for
metric hue and chroma are as, for example, defined in the CIELAB formula
(see Chapter 6).

4.6 CONFIRMATION OF THREE ATTRIBUTES

The attributes hue, lightness, and chroma for surface colors in simplified
viewing conditions have received confirmation from multidimensional scaling
experiments. In 1988 Indow reported on the results of nineteen multidimen-
sional difference scaling studies using Munsell color chips. Because of diffi-
culties in comparing chromatic differences of colors with greatly differing hue
scaling was only done in overlapping, comparatively small regions of the
Munsell chromatic plane. Indow drew the following key conclusions:

1. In a three-dimensional euclidean space the points representing Munsell

colors form layers according to the order of Munsell value.
2. In each of the planes colors of the same Munsell hue are located along
line segments in the order of Munsell chroma; all lines converge at a
single point in the center, corresponding to the gray of the same Munsell
value. The circular order of the radial lines agrees with the sequence of
Munsell hue.
3. Interpoint euclidean distances between colors are closely related to
scaled perceptual distances that were used as the data in the multidi-
mensional scaling experiment.
4. Chroma and value are clearly orthogonal.
5. There are many irregularities in the resulting structure, the main ones
being anomaly of hue spacing in sectors B to P, and the first chroma step
from neutral is always larger than successive steps.

The result of one multidimensional scaling analysis of Munsell colors at

different value levels is shown in Fig. 4-4. Taken together, the findings help
support the Munsell system as a systematic perceptual arrangement of colors
and the psychological validity of the three attributes.
The perceptual significance of the three Munsell attributes is also indirectly
supported by subjecting the reflectance functions of the Munsell chips to
dimensionality reduction using a neural network (Usui et al., 1992). The
 CONFIRMATION OF THREE ATTRIBUTES 135

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 4-4 Multidimensional scaling analysis of Munsell colors involving samples at four differ-
ent value levels (From Indow, 1988). The results are shown in an elliptic space. Colors of
constant hue are connected with approximately radial lines, colors of equal chroma fall on
approximate circles.

reflectance data of 1596 chips were subjected to a five-layer wineglass type of

neural network. Of the three middle-layer units one was found to approxi-
mately correspond to value and the other two roughly to opponent responses
with red-green and yellow-blue orientation. Lines of constant hue are approx-
imately radial and lines of constant chroma jaggedly oval but placing the hues
in proper ordinal order. When reducing the network to two internal layers,
yellow and blue colors overlap. On the other hand, of four layers one was
found to be redundant, two layers of the four being highly negatively corre-
lated (r = -0.991). The most efficient representation involves three internal
layers and reconstructs the Munsell color solid roughly as in the X, Y, Z tris-
timulus space (Fig. 4-5; compare to Fig. 5-30).
Because of the diurnal cycle, to whose effects we are exposed already in the
womb, the ideas of bright (or light) and dark are deeply embedded in us. In
their simplest form they are achromatic experiences, and all but congenitally
blind people know them. As to chromatic colors, the preeminent fact is hue.
This is also indicated by our monolexemic names for what we regard as the
most significant colors. They generally refer to hue. Level 2 of the Universal
Color Language consists of 26 hue-related names (Kelly and Judd, 1976). On
136 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

Fig. 4-5 Reconstruction of the Munsell color solid from 1280 reflectance functions of Munsell
chips with a five-layer neural network (From Usui et al., 1992). Colors of constant value fall
approximately on planes of constant unit 3 values. Constant hue colors fall on approximately
radial lines and colors of constant chroma are connected by jagged ellipsoidal contours. The
arrangement is similar to one of Munsell colors in the CIE X, Y, Z space.

the other hand, the number of modifiers added in level 3 that describe light-
ness and chroma (e.g., light, strong, vivid, dark, and deep) is only 13. These,
together with the 26 hue terms of level 2 and the achromatic terms form the
267 terms of level 3. From this and other facts it is evident that hue is the
primary attribute of our normal color experiences. In evolutionary terms light
and dark are older, but they encompass one dimension. Hue, on the other
hand, has, in a manner of speaking, four dimensions and therefore provides a
much larger range of experiences. The general idea of a third attribute of
object colors, now named chroma, developed beginning in the nineteenth
century with people professionally involved with color, namely painters. In the
general population today, even in well-educated segments, it is not yet a com-
monly understood concept.
The attributes of the Hering-NCS systems are claimed to be intuitive by
their creators and to be introspectively determinable with a high degree of
precision. The author is not aware of an extensive independent test of this
claim. The clear expectation is that in a multidimensional scaling experiment
based on uniformity of perceived differences NCS color chips would be placed
into a Munsell type color solid.
 NEURAL CORRELATES OF COLOR ATTRIBUTES 137

4.7 CONTRAST VERSUS SIMILITUDE

Given an arrangement of two test fields against a surround sending different

spectral power distributions to our eyes, the color experiences resulting from
the test fields depend on the spatial proximity of those fields. There are two
results possible: either the perceived difference between the test fields
becomes larger if the fields are adjacent compared with widely separated or
it becomes smaller. In the former case we speak of contrast, in the latter of
similitude (also known as the spreading effect; Bezold, 1874). Such effects are
usually interpreted as due to lateral interaction between retinal cells (but see
Purves and Lotto, 2002). As a result the relationship between stimulus and
resulting perception depends on the composition of the visual field.
Contrast has a somewhat different meaning in photography or printing and
is defined as the difference between the lightest and the darkest area in the
image expressed perceptually or in terms of physical measurements. Aside
from lightness contrast there is also chromatic contrast.
Discovery and quantification of contrast as a function of spectral energy is
a key aspect of operation of our visual system. As will be shown in Chapter 5,
perceptions of white and black do not depend on absolute values of spectral
power but are related according to complex, not yet fully understood rules.
Similarly it appears that colors are assigned by the visual system to fields of
given spectral power, and not by that power but by the relationship of that
power to that of neighboring fields according to rules not yet fully understood.
Colors may be said to exist only as contrasts since in a so-called (extreme)
Ganzfeld where our eyes are individually exposed to uniform light from all
directions color experience fades quickly regardless of the spectral power dis-
tribution of the light. While contrast effect rules are comparatively simple in
case of two fields only, the more articulated the visual field is, the more
complex the rules become. Contrast effects resulting in colored shadows were
described in the eighteenth century by the Count of Rumford and others.
Chevreul famously developed empirical rules of contrast in the nineteenth
century. Contrast effects are temporally dependent. When viewing two con-
trasting fields in a neutral surround, the simultaneous contrast is immediately
apparent. When the two fields are replaced after some time by the surround
color, their images remain still visible, however, with appearances usually
changed according to the rules of successive contrast.

4.8 NEURAL CORRELATES OF COLOR ATTRIBUTES

Color attributes are phenomenological entities that have as yet no firm basis
in neurology. Current answers to the question of neural correlates of color
attributes are speculative. According to the opponent color theory, there are
six primary object color perceptions: unique yellow, red, blue, green, and white
and black. A chromatic object color stimulus invokes one or two of the unique
138 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

hues, and perhaps white and/or black. It is not yet known what the neural cor-
relates of the unique hue perceptions are. Known opponent color cells in the
lateral geniculate nucleus (LGN) and beyond in the cortical visual system do
not have outputs correlated with unique hue perceptions, as discussed in
Chapter 5 (see also Webster et al., 2000; Valberg, 2001). It appears less and less
likely that there are two simple subtractive chromatic opponent color systems
as proposed by Hering. Multiple hue detection and discrimination systems
have been proposed as an alternative to a two-component opponent color
process (e.g., see Krauskopf, 1999). There is a large amount of literature relat-
ing to the operation of a neural opponent color system, a subject outside the
scope of this text (relevant citations are found in Gegenfurtner and Sharpe,
1999).
Brightness-related signals are transferred from the retina through at least
two different neural pathways. Brightness and lightness perception is very
complex, for example, as the work by A. L. Gilchrist and co-workers shows
(Gilchrist, 1994; Gilchrist et al., 1999), and the Munsell value and similar light-
ness functions are only simple approximations. Recently Lotto and Purves
(1999) have argued that brightness/lightness perception of fields may be
guided by our past experiences as a species and as individuals, and interpre-
tation of clues in an image in regard to its likely illumination.
Hue and brightness/lightness perceived in contrast to a surround are
affected by a complex set of adaptations whose neural machinery is not well
known. As will be seen in Chapter 5, the neural machinery of hue and chroma
detection and discrimination is also not known, but signs point to a complex
system.

4.9 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF

COLOR ATTRIBUTES

This section contains a discussion of various scaling attempts of the three fun-
damental color attributes hue, chroma, and lightness. The section on hue
includes the issue of the number of uniform hue difference steps between
unique hues at a given level of chroma. The chroma section includes a dis-
cussion of the determination of the magnitude of the first steps from gray, pre-
sumably an indication of the coloring power of the unique hues and their
combinations. Lightness scaling is briefly touched upon, but most of the dis-
cussion of lightness scaling is postponed until Chapter 5.

Hue

Newton’s hue circle was shown to be a parsimonious solution to representing

the common desaturation effect on all spectral hues. When scaling hue differ-
ences of fourteen spectral colors and performing multidimensional analysis
of the results, Shepard (1962) obtained data points that are fitted with good
 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 139

accuracy by a circle, with a considerable space between spectral violet and red,
as one would expect (see Fig. 3-6). With availability of purple hues in object
colors the hue circle is naturally closed, as Hering has pointed out.
Unlike lightness and chroma, hue cannot easily be envisaged by reference
to simple numbers. To have a concept of hues based on hue angles as deter-
mined in a given system is too abstract to be of practical use. One can have a
mental image of what a value 6 gray scale chip looks like, and it may even be
possible to have a mental image of a given hue identified by Munsell hue name
at chroma 8, but this is no longer possible with any reasonable accuracy for a
hue with a hue angle of 165° unless intensive training has taken place.
In general, it is easier to assign unique hue, chroma, and lightness values to
a given color stimulus than to mentally synthesize it based on such numbers.
Without extensive practical experience I can more easily estimate the content
of yellow, red, and gray in a brown than create a mental image of a brown
from numbers of yellow, red, and gray given to me. Our color names are rel-
atively vague and most have considerable extension around focal points. Only
the unique hues give us a comparatively solid mental reference point (however
individually variable as to the required stimulus, as will be seen below). Mid-
points between unique hues also can be fixed relatively well, though less well
than unique hues. On the other hand, “purplish pink” or “turquoise” can apply
to five or more (individually varying) Munsell 40-hue steps. In general, color
names are least well defined in the yellow-green segment, probably because
of the lack of widely present exemplars during the time period when color
terms developed.
As mentioned, division of Newton’s hue circle is proportional to the inter-
vals between the musical sounds of an octave, thereby implying a relationship
between sound and color (Newton, 1704). Newton’s circle did not explicitly
include extraspectral (purple) hues. Grassmann’s circle is essentially identical
to Newton’s. Helmholtz’s circle, at highest saturation, includes extraspectral
hues and is divided into ten equal segments with the middle color of each
segment diametrically opposed to its complementary color. Hue circles based
on complementary hues being diametrically opposed were common after
Helmholtz. However, there is no simple information on which to base the
placing of complementary diametrical lines relative to each other. In both his
earlier sphere and later cone models Wundt placed complementary hues oppo-
site. In case of the sphere model they were placed in such a manner that the
largest segments are occupied by yellow and blue. Rood studied this matter
and concluded that there was no objective way of placing the diametrical hue
lines relative to each other. His ten-hue chromatic circle has uneven division
between the hue lines and is apparently based on a placement using distances
along the spectrum scale with arbitrary placement of the nonspectral reds and
purples (Rood, 1879). There seem to be only two methods by which to place
hues with perceptually meaningful distances into a circle.
Hering was, as mentioned, the first basing a hue diagram on four unique
hues. His psychological hue circle (Fig. 2-32) has conceptually equal percep-
140 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

tual increments/decrements of unique hues. As shown below, hue coefficients

describing the relative perceptual amounts of unique hues in Hering’s system
form by definition an ¥ pattern. The NCS system represents a practical imple-
mentation of Hering’s hue circle.
Munsell’s approach to scaling a hue circle was based on uniformity of the
size of hue differences along the hue circuit. As a result complementary colors
as expressed by optimal colors of Munsell Renotations in the CIE chromatic-
ity diagram do not plot diametrically opposed in the Munsell psychological
diagram. We can conclude from this that chromatic circles based on comple-
mentary colors are not uniform in terms of perceived hue differences. As we
will see below, the Hering/NCS circle also is not uniform in terms of perceived
hue differences, indicating that constancy of perceived unique hue increment
does not equal uniform size of perceived hue difference in the four quadrants.
An interesting issue in connection with hue scaling is the psychophysical
identity of the four unique hues, given an achromatic surround. Historically
these have been determined in the spectrum, using optical equipment. Indi-
vidual results have varied to a surprising degree (e.g., see Ayama, Nakatsue,
and Kaiser, 1987). For example, the wavelength of unique green has been
determined in various experiments as anywhere from 488 to 561 nm. Similar
results have more recently been reported by Webster and co-workers (2000).
Unique hues can also be determined, with considerably reduced variability,
using object color samples (Hård, Sivik, and Tonnquist, 1997; Kuehni, 2001a).
The mean choices of 40 observers that directly selected their unique hues from
arrays of Munsell hues were found to be 3.0R, 3.5Y, 2.5BG, 2.75PB. Indow,
based on his principal hue scaling method, obtained the following locations:
3.75R, 5Y, 6G, 3PB (Indow, 1999b). Note the discrepancy in the green unique
hue. In a recent, as yet unpublished, study using Munsell chip arrays with 75
observers average unique green was found to be 7.0G. Webster’s group
(Webster et al., 2000) found green to have the largest individual variation
among unique hues. This was also the case for Kuehni’s original experiment
as well as the experiment with 75 observers. It is evident from results with
Munsell chips that the range of dominant wavelengths in such experiments is
much narrower than the reported range of spectral wavelengths for determi-
nations using optical equipment. Recently it was shown that there is no rela-
tionship, as previously surmised, between the ratio of L and M cones in an
individual’s retinas and the perception of unique yellow (Yamauchi et al.,
2002). There is currently no model with good explanatory power for the rela-
tionship between stimulus and perceived unique hues.
Conceptually the geometrical image of a perceptually uniform hue circle of
constant chroma is a perfect circle with each hue placed at uniform hue angle
increments around it (Fig. 4-6). The location of each hue can be described in
terms of ordinate (b) and abscissa (a) values relative to the center of the circle.
Logically it might be expected that these values represent the content of unique
hues in each hue. But, as we will see, this is not the case if hue steps are per-
ceptually uniform. When plotting these absolute coordinates against the hue
 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 141

Fig. 4-6 Idealized uniform chroma circle with forty colors representing identical hue differences
in an a, b opponent color diagram.

angle, sine wave curves are obtained (Fig. 4-7).When these are converted to rel-
ative values (percentages, called hue coefficients) ¥ pattern functions result
(Fig. 4-8). An ¥ pattern in agreement with Fig. 4-8 is also obtained when plot-
ting the forty NCS hue steps, conceptually based on uniform changes in unique
hue components. In its psychological diagram the NCS atlas hue circle plots as
forty equal hue angle increments of 9°. The Munsell atlas hue steps, involving
constant hue differences, plot in the same way. However, this diagram is not
based on the four unique hues and has no preferred axis system. In the stan-
dard presentation of the Munsell hue circle the x axis is formed by hues
10PB and 10Y, the y axis by 5BG and 5R (as can be seen in Fig. 4-11 below).All
four are, to a greater or lesser extent, hues that are mixed in terms of unique
hues. But while a step along the NCS hue circle means a 10% change in unique
hue content, a step along the Munsell hue circle means a constant hue differ-
ence 1/40th of the total hue circuit. As a result the two diagrams are quite
different.
The hue scales of the Munsell system up to and including the Munsell Re-
renotations are briefly described in Chapter 7.Various hue scaling experiments
142 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

Fig. 4-7 Plot of a and b values of the ideal color circle of Fig. 4-6 as a function of hue angle.
The points form sinusoidal curves.

Fig. 4-8 Hue coefficients (relative a and b values) as a function of hue angle, calculated from
the absolute values of Fig. 4-6.
 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 143

using Munsell Book of Color chips have been performed by Indow and his
colleagues. In 1972 Indow and Ohsumi published the results of multidimen-
sional scaling involving 60 samples. The corresponding hue circle was consid-
erably distorted compared to the Munsell circle. The hue angle between 5PB
and 5P, 36° in the psychological diagram, was found to be 80°. In a later more
extensive multidimensional scaling experiment involving 176 Munsell chips, a
hue circle much more in geometrical agreement with the conceptual Munsell
circle was found (Indow and Aoki, 1983). The difference, in the most recent
analysis of the results (Indow, 1988), between 5PB and 5P at the highest
chroma level is 45°. In a later study Indow and co-workers had observers assess
the principal hue components in Munsell chips at different chroma and value
levels (Indow, 1999b). They had observers mark on a paper scale the estimated
amount of achromatic (gray) color in the total color. Next observers deter-
mined the content of one or two principal hues in the color and marked them
separately. Average absolute principal hue components at chroma 8 and value
6 are illustrated in Fig. 4-9; There is a reasonable resemblance to Fig. 4-8;
however, yellowness is undervalued compared to the other three principal hue
components (presumably because observers disagreed on what represents
100% yellow). The resulting implicit, considerably distorted, hue circle is illus-
trated in Fig. 4-10. Implied chroma varies substantially, as do the hue angle dif-
ferences. Somewhat different results have been obtained at other levels of
chroma and value. The changes in absolute hue components are more linear

Fig. 4-9 Average absolute principal hue components of Munsell colors at chroma 8 and value
6 (after Indow, 1999b). The curves have a resemblance to those of Fig. 4-8. The biggest devi-
ation is for yellow, presumably because the observers could not agree on which color repre-
sents unique yellow.
144 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

Fig. 4-10 Hue circle derived from the absolute hue components of Fig. 4-9.

than sinusoidal between unique hues, indicating that the observers judged the
chromatic content of mixed hue colors lower than that of unique hue colors,
most severely so for 1 : 1 mixes (compare Fig. 4-10 and Fig. 4-6).

Number of Uniform Steps between Unique Hues

In the hue circle of his atlas Hesselgren placed different numbers of equal-
sized hue steps between his unique hues (Hesselgren, 1952). There are eight
steps from blue to red, six from red to yellow, and five each from yellow to
green and from green to blue.
Somewhat different results were obtained by Kuehni who investigated the
perceptual distance between unique hues. Six observers scaled the Munsell
hue circle at value 6, chroma 8 by using the difference between 5PB and 10B
as the reference difference. The five Munsell 100-hue steps difference of the
reference pair was seen as equal in difference to between 3.0 and 5.8 steps
around the hue circle. On average, a total of 22 (rather than the expected 20)
hue steps of the size of the reference step was found. In the four sectors
between the unique hues the number of equal sized steps was found to be dif-
ferent: R–Y: 5; Y–G: 6; G–B: 4; B–R: 7 (Kuehni, 1999).
When plotting the average unique hues of Kuehni’s 40 observers (green
adjusted to the results of the 75 observers) on the ideal psychological Munsell
hue circle (Fig. 4-11), the yellow and blue unique hues fall on a straight line
passing through the center while red and green do not. The lines do not sep-
 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 145

Fig. 4-11 Ideal Munsell hue circle with the average unique hues determined by the author
(and an additional experiment for unique green) represented by heavy full lines.

arate the hue circle into four equal segments. Rounded to the nearest half-
step on the 40-step Munsell hue scale the number of steps in each sector is as
follows: R–Y: 8.0; Y–G: 9.5; G–B: 10.5; B-R: 12.0. If there are four primary hues
and all other hues are mixtures of two of them, one might, sensibly and logi-
cally expect the primary hues to fall on the axes of a Cartesian diagram. Only
then, as we saw above, can the content of primary hues in any given hue be
expressed in terms of the axis coordinates. The Munsell system indicates,
however, that in such a diagram there is a nonproportional relationship
between hue angle and perceived hue difference. For reasons presented in the
next chapter we cannot absolutely rely on the Munsell hue differences at a
given chroma level having all exactly identical perceived magnitude and addi-
tional constant hue difference scaling work is desirable.
A nearly identical figure results when locating the same unique hues in
Indow and Aoki’s multidimensional euclidean scaling grid of the Munsell hue
plane (fig. 6 in Indow 1988). The conclusion here also is that the number of
unit-sized hue differences in the four sectors between unique hues differs.
In addition to the experimental results above the following relevant data
are available:

1. Based on published information for the DIN color system, it is possible

to translate the same unique hues into that system. The number of
146 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

TABLE 4-1 Comparison of hue angle segments in the

psychological chromatic diagram between unique hues
from different psychological experiments
Data set Angle of segment, Degrees
R–Y Y–G G–B B–R
Munsell Renotations 71 84 96 109
Hesselgren 90 75 75 120
DIN system 90 75 60 135
Newhall experiment 63 89 82 126
OSA-UCS 66 83 85 126
Kuehni experiment 82 98 65 115
Qiao et al. experiment 76 76 86 122
Berns function 65 82 87 126
Mean 75.4 82.8 79.5 122.4
CV, % 14.5 9.6 15.2 6.5

uniform hue steps between the unique hues in the DIN system is R–Y:
6; Y–G: 5; G–B: 4; B–R: 9.
2. As part of a preliminary effort during the development of OSA-UCS
Newhall scaled a constant chroma circle at value 6 into 40 visually
equally sized hue differences (102 observers, 142 pairs, 14,484 judgments;
Judd, 1965). Unique hues have been plotted into an optimized model of
the Newhall results.
3. They also have been plotted into the g, j diagram of the OSA-UCS
system, regarding it as an optimized psychophysical model of the psy-
chological diagram.
4. A hue difference scaling experiment for the purposes of establishing a
hue difference weighting function has been performed by Y. Qiao et al.
in 1998. The resulting CIELAB weighting function can be integrated and
the number of equal hue steps between the unique hues determined.
5. R. S. Berns has proposed a somewhat modified hue function, taking into
account suprathreshold small color difference data (Berns, 2001).

In Table 4-1 the hue angle segments in the psychological chromatic diagram
between the four unique hues of the various experiments are compared
(Kuehni, 2001e). As Table 4-1 shows, there is considerable variation in the
results, but the trend is uniform (except for the yellow sector in the Hessel-
gren and DIN systems). The variation coefficients (CV) indicate greatest
agreement for the R–B sector and significant disagreement in the R–Y and
G–B sectors. Additional high-reliability hue-scaling experiments to provide a
clearer picture of this important issue are very desirable.

In addition a linear model was fitted to the MacAdam color-matching error

 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 147

ellipses resulting in a coefficient of variation of 15% for 100 color differences

(four per ellipse; Kuehni 2001d). By this formula the number of hue color
matching error steps between the average unique hues was calculated with the
following results: R–Y: 52; Y–G: 77; G–B: 72; B–R: 127. The trend is the same
as that shown in Table 4-1, but the differences in the segments (after scaling
the numbers for a total of 360) are more distinct.
In the NCS system unique hues form by definition equal sized quadrants
in their hue circle. The NCS unique hues are well within the observer vari-
ability of those of Kuehni’s unique hue determination experiment (and the
additional experiment on unique green). As mentioned earlier, the implica-
tion is that the psychological unit differences are not the same in the differ-
ent quadrants of that system. This has been mentioned by Tonnquist (1966).
It follows that equal numerical changes in perceived content of unique hues
do not result in equal perceived hue differences in the four segments.

Chroma
While hue has four comparatively easy to locate psychological markers, the
unique hues, chroma has none. Different chroma scales (the results of differ-
ent experiments) can therefore only be compared in terms of physical or
psychophysical scales. Such comparisons will be made in Chapter 6.
Chroma steps have been investigated extensively as part of the develop-
ment of the Munsell system as described in Chapter 7 and the results codified
in the Munsell Renotations. The Renotations represent a considerable change
of chroma scaling compared to the 1915 Atlas and the 1929 Book of Color.
Few explicit studies of chroma scaling have appeared since then. In 1957, in
preparation for the development of the Optical Society of America Uniform
Color Scales (OSA-UCS), Judd, Nickerson, and Nimeroff determined a
constant chroma circle at value 6 and chroma 8 (384 chroma differences, 60
observers, 23,040 observations; Judd 1965). As Chapter 5 will show, it differs
significantly from the Renotation constant chroma circle. Implicit chroma
scales along the chromatic axes were also developed for the OSA-UCS system.
The four chroma scales, when compared in a common psychophysical system,
are found to differ considerably, as shown in Chapter 5.
In the DIN system saturation rather than chroma was scaled. The constant
saturation contours of that system in the CIE chromaticity diagram differ
significantly from the constant chroma contours of the Munsell system. DIN
saturation is independent of lightness and the relationship between chroma
and DIN saturation differs as a function of lightness.
NCS chromaticness is different from Munsell chroma. There has been no
recognition in that system of the varying chromatic power of optimal colors
as all implied full colors arbitrarily have been set at chromaticness 100. When
transforming Munsell colors of constant chroma to NCS notation the result-
ing chromaticness varies with hue in roughly sinusoidal fashion, as one would
expect from the definition of NCS chromaticness.
148 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

Chroma scales are also an implied result of Indow’s principal hue and chro-
matic versus achromatic color component determination as well as of his mul-
tidimensional scaling work. The scales from the two experimental efforts differ
significantly. In the principal hue component work the implied chroma scales
are often severely compressed at higher chroma levels, indicating the difficulty
observers had in assessing the relative amount of achromatic and chromatic
components (see Fig. 4-9). There was also some compression of the chroma
scale at high chroma levels in the multidimensional scaling experiment, when
compared to the Munsell system. However, when chroma scales of the Book
of Color atlas are viewed, the higher chroma steps do not generally appear
somewhat smaller than those at lower chroma. Indow also appeared to accept
the Munsell chroma scales as essentially valid since he fitted a Riemannian
model to his MDS data to improve the evenness in regard to chroma. The
results raise doubt that chroma can be accurately assessed by determination
of principal hue and achromatic components in a color or by multidimensional
scaling.
Chromaticness scaling used in all cases, for sensible reasons, an achromatic
surround. Chromatic surrounds change chromaticness scaling. However, this
subject is outside the scope of this text and an issue for general appearance
modeling. As will be argued later in this text, chromaticness depends on the
magnitude of the unit chromaticness difference used to scale it (see also the
related comment in the next section).
Chromatic thresholds, the first steps from the neutral point, have been
studied since the 1920s. Typical saturation thresholds (amount of spectral light
to be added to white light to result in a just perceptible color) as determined
in 1938 by I. G. Priest and F. G. Brickwedde are shown in Fig. 4-12. Results by
other investigators differ in detail but agree in the general features. Such
thresholds are, presumably, indicators of chromatic strength of spectral colors.
Evans defined spectral chromatic strength as the factor by which the lumi-
nance of a stimulus has to be multiplied in a brilliance match to equal the bril-
liance of the achromatic surround (Evans, 1974). Recall that a brilliance match
against the surround is achieved when the stimulus field looks neither grayish
nor fluorent against the surround. In 1968 Evans and Swenholt investigated
chromatic strength using the Munsell system. In plotting the reciprocal CIE
colorimetric purity against the dominant (or complementary) wavelength for
chroma circles 2, 4, 6, 8, and 16 at value 5, they obtained curves essentially
parallel to each other and reciprocal to the saturation threshold curve (see
Fig. 4-13). These curves provide another kind of support for the salience of
the chroma attribute.
In an experiment with 35 observers Kuehni investigated the relative sizes
of the first steps from gray of the colors nearest to the respective axes in the
a*, b* diagram in the Munsell as well as the OSA-UCS systems (Kuehni,
2000c). The results indicated that the average observer judged, in the Munsell
system at value 6/chroma 8, the first step toward yellow to be about 2.5 times
the magnitude of the first step toward blue. In the red/green direction the
 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 149

Fig. 4-12 Spectral saturation thresholds determined in 1938 by Priest and Brickwedde. From
Osgood (1953).

green step was seen as about 10% larger than the red step. For OSA-UCS
the first step toward yellow was seen as about 20% larger than the first step
toward blue. In that system the first steps toward red and green were seen as
of about equal size. Somewhat different results have been obtained by Indow
(2001).
The results of Priest and Brickwedde and many later researchers were
obtained monocularly with optical apparatus. M. De Matiello and co-workers
investigated the difference between monocular and binocular saturation dis-
crimination at the threshold level as well as for suprathreshold saturation dif-
ferences (2001). They found the changes in colorimetric purity for threshold
steps to be considerably larger for the first step from gray than for the first
step from the spectral color. For suprathreshold saturation differences they
found power relationships varying by wavelength with exponents from 0.61 to
0.97. Interestingly they found exponents of binocularly determined data to be
lower than those obtained for monocular data.

Relationship between Hue and Chroma Differences

This matter created considerable discussion in the 1950s and 1960s (Judd,
1969).As mentioned previously, in 1936 Nickerson developed the first formula,
on a psychological basis, that could result in predictions of the perceptual
150 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

Fig. 4-13 Reciprocal CIE colorimetric purity as a function of dominant wavelength for Munsell
colors at five levels of chroma (represented by different symbols) and value 6. From Evans and
Swenholdt (1968).

magnitude of color differences, the Index of Fading. It was based on the

Munsell system and reads as follows:

Ê C ˆ(
I= 2 DH ) + 6 DV + 3DC , (4-1)
Ë 5¯

where I is the index value, C is Munsell chroma, V is value, and H is hue

expressed in terms of the Munsell 100-hue circle. For two colors being com-
pared, perceptual differences according to this formula are equal to the simple
sum of weighted differences between the index values. Note that this formula
predicts hue differences between colors of neighboring Munsell hues to grow
linearly in magnitude as a function of chroma.
As Judd showed, for colors of equal V one hue step is, according to the
Nickerson formula, equal to 2C/15 = 0.133C chroma steps. When expressing
the total hue angle of all 100 hue differences the result is found to be 13.33C.
This is slightly more than twice the theoretical total hue angle of a circle that
in a euclidean system is 6.28C.While initially skeptical of this result and believ-
ing it to be due to experimental error, Judd later found support for what he
 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 151

called the “superimportance of hue” in experiments involving large hue dif-

ferences. If the Nickerson formula is factual, perceptual color space is not
euclidean. Judd used a crinkled fan model to account for hue superimportance
(see Fig. 7-15).
Hue superimportance was also detected in the experimental basis data for
the OSA-UCS system. In 1964 the Committee on Uniform Color Scales of the
Optical Society of America completed a basic experiment eventually result-
ing in the OSA-UCS scales. Forty-three samples were viewed in 102 pairs by
71 observers and compared for size of difference against other pairs. After
psychometric scale values were calculated, the correlation between these
and various formulas was determined. The correlation coefficient obtained, for
example, for the Glasser cube root formula was 0.34 while the Nickerson
formula resulted in a value of 0.61. When, instead of sums, the square root of
the sum of the squares of the component differences was calculated, the cor-
relation increased to 0.65 (perhaps merely a fortuitous result). An optimal
value of 0.80 was obtained with (1) revised Munsell spacings in regard to hue
and chroma and (2) use of the following formula, which derives from a formula
developed earlier by Godlove as a description of the ideal Munsell system in
a euclidean space in place of the Nickerson formula:7

0.5
DE = {[ f f (C C )
g h 1 2
0.5
]
2 2
DH + (DC ) + (4 DV )
2
} , (4-2)

where C1 and C2 are the Munsell chromas of the two colors compared, DH is
the hue difference in Munsell 100-hue steps, DV is the difference in Munsell
value, and

0.5
[2( 1 - cos 3.6 ∞ DH )]
fg = ,
DH
2 - k + 4(k - 1)
fh = ,
3 - cos 3.6 ∞ DH

where k is the hue superweight factor with a value from 1 to 2. If the value is
1, then fh = 1. For optimal correlation the value of k was selected as 1.7 (Judd
and Nickerson, 1967), this being indicative of hue superimportance.This exper-
iment supported the idea of a relationship between perceptual hue differences
and chroma differences that does not fit into a euclidean system. This matter
is further discussed in Chapters 7 and 8.

Lightness
The first “qualitative” verbal gray scale, seemingly, is the one by Forsius, with
seven grades including black and white (see Chapter 2). Glisson described
quantitatively (in terms of weights of black and white pigment to be used) a
152 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

24-grade gray scale that is reasonably uniform through the middle portion.
Mayer envisaged 23 grades of different lightness in his double pyramid, while
Lambert reduced that to 8 grades. Runge’s color sphere has a 10-grade gray
scale.
In 1729 the French mathematician Bouguer published Essai d’optique sur
la gradation de la lumière (Optical treatise on the gradation of light) in which
he described experiments using shadows of rods made by the light of two
candles on a white screen. He found that the distance ratio of the two candles
needed to be about 8 : 1 to result in a just noticeable difference between the
two shadows. Thus he initiated the study of brightness thresholds. In 1845 his
compatriot V. Masson described a white disk on which a black radial line
segment was inscribed. When spinning the disk the line segment darkened a
ring of the disk (the spinning disk method was invented by Pieter van Muss-
chenbroek in 17688). The width of the line segment as a fraction of the total
disk circumference is an indicator of the fractional change in luminosity
required to see a difference. Masson found that the ratio, depending on cir-
cumstances of illumination and viewing distance, was between 1/50 to 1/120.
In 1850 the French astronomer D. F. Arago (1786–1853) repeated the shadow
experiment with improved equipment and found a ratio of 1/133. Masson’s
experiment was repeated in 1858 by Fechner and Volkmann, who found a ratio
of 1/100 and by Helmholtz in 1860 who, under optimal conditions, could see a
ratio of 1 : 167. In 1888/9 König and Brodhun used Masson’s technique to
determine brightness as well as chromatic thresholds.
Attacking the problem of a uniform gray scale from another angle Plateau
in 1872 asked several painters to paint a color halfway between black and
white (as described in Chapter 2), thus performing an equisection. Indepen-
dently Delboeuf developed a detailed uniform gray scale using successive
equisection, as well as by adjusting black and white disk segments to obtain
perceptually equal steps.
In 1874 Hering, as part of his psychological color triangle, described a
“nuanced” series of grays from black to white that was to have uniform incre-
ments/decrements of blackness and whiteness. Ebbinhaus described in 1887
an eight-grade gray scale (see Chapter 5). In 1899 Munsell began to develop
his color order system of which an eleven-grade (ten-step) value scale was an
integral part. The Munsell value scale underwent development until the 1930s
and was finalized in the Munsell Renotations. In the final experiments gray
scales were visually measured against a white, gray, and black background.
They were found to differ, and the results were averaged into the value scale
(see Chapter 5, Section 5.7). Ostwald selected a logarithmic relationship
between the stimulus and the response for his gray scale. In the 1960s Kaneko
and Takasaki confirmed and further quantified W. Schönfelder’s finding that
color differences between two stimuli are perceived best if the surround is
intermediate between the two stimuli. This makes a gray scale and its steps
dependent on the surround. Surround dependence had been found in 1922 by
Adams and Cobb and was also investigated by Evans. The effect was included
 PSYCHOLOGICAL (PERCEPTUAL) SCALING OF COLOR ATTRIBUTES 153

into the definition of the OSA-UCS lightness scale, otherwise based on the
Munsell value scale. The NCS whiteness/blackness scale is derived from
average judgments of the content of whiteness and blackness in gray samples.
The atlas scale has eleven grades (including black and white) resulting in ten
steps, the numerical scale has 100 steps.
From the beginning with Bouguer the development of lightness scales has
run parallel with the development in photometry. While work with the Masson
disk or Delboeuf’s equisection method and others did not explicitly require
photometry, their results were best interpretable in quantitative terms when
compared to the corresponding stimulus strength. The development of light-
ness scales is therefore further discussed in Chapter 5.
A fundamental issue involving lightness is how brightness and lightness are
connected. As has been demonstrated by many researchers over the last 100
years, a source of achromatic color with any given luminance value can,
depending on the conditions it is seen in, be perceived as any grade of gray
from white to black. The question arises how the human visual system decides
to assign a given lightness value to a given luminance value. Among the early
researchers investigating this matter was D. Katz who was a student of G. E.
Müller. In 1909 he published a book-length paper, Die Erscheinungsweisen der
Farben und ihre Beinflussung durch die individuelle Erfahrung (translated in
1935 as The world of colour) in which he discussed his investigations of dif-
ferent appearance modes of colors, including the relationship of brightness,
lightness, and luminance. In a dramatic experiment Gelb showed in 1929 that
if a black piece of paper is suspended in midair and illuminated with a strong
beam of white light, its appearance is white. If a paper of higher luminous
reflectance than that of the black paper is placed next to it, the original paper
no longer appears white but a shade of gray or black, depending on the lumi-
nous reflectance of the second paper. Thus lightness perception is dependent
on the relative luminous reflectance of adjacent fields. A few years later gestalt
psychology theorists postulated that the visual system compares the luminance
of a target against the weighted average of the luminance of the total scene.
In 1948 H. Wallach proposed that lightness is decided by the ratio of the lumi-
nances of two adjacent fields. The so-called intrinsic image model of lightness
was developed in the last twenty years. It had been found that when a gray
paper is successively viewed against backgrounds of different lightness, the
ratio of luminous reflectances changes significantly but the perceived lightness
of the paper changes little. Similarly there is considerable lightness constancy
when viewing the arrangement of paper and background under different kinds
of illumination. The intrinsic image model analyzes the image as it appears on
the retina according to three components: surface reflectance, illumination,
and three-dimensional form clues. However, significant disagreements
between the predictions of theory and the experimental results remained in
certain situations. As Gilchrist and colleagues (1999) describe it, a fundamen-
tal question is that of the anchoring problem. It relates to the question “where
to locate the range of luminance values (in a given situation) on the scale of
154 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

perceived gray shades.” Once it is solved, the next question is “how the range
of luminance values is distributed on the scale of perceived gray shades.” Many
researchers have contributed to the current answers: in simple images (two
fields of different luminance filling the entire visual space) the highest lumi-
nance is seen as white unless the relative area of the darker field is more than
half of the total field. Then an area rule applies. As the darker side grows in
relative size, it appears lighter and lighter, making the smaller, lighter area
appear more and more white, eventually fluorent and then self-luminous.
In more complex images Gilchrist and his group propose to separate the
image into components, called frameworks, that belong together according to
gestalt principles (an exact definition of framework is difficult, so coplanarity
of the elements appears to be a good start). For each local framework the
anchoring and area rules of the simple situation approximately apply. The
entire visual field is called the global framework. Within a local framework a
veridical, or 1 : 1 scaling between lightness and luminance, is found to apply if
the luminance ratio between white and black is less than 30 : 1. If it is larger,
some luminance scale compression occurs. Gilchrist et al. have tested their
model on many classical lightness test results and “illusions,” such as the
Benary effect (1924) or White’s illusion (White, 1981) and found good explana-
tory power. In cases where the information from local frameworks appears
contradictory, Gilchrist et al. believe that our visual system makes a compro-
mise between the results or alternately displays one or the other. It is evident
that much work remains to be done.

4.10 PERCEPTION OF COLOR DIFFERENCES

A formula that mathematically describes the relationship between attribute

component differences and total perceived difference is Nickerson’s formula
of 1936 encountered above and expressed in terms of the attributes as defined
in the Munsell system. The investigation by Bellamy and Newhall (1942)
showed that the definition of the formula depends on the size of the differ-
ences; that is to say the quantitative relationship between the Munsell attrib-
utes changed significantly when tested at the JND level.
In current practice the results of small color difference calculation are
usually expressed in terms of total color difference and its hue, chroma, and
lightness difference components. The visual small color difference data have
in all cases been determined as total differences. The split of the total differ-
ence into components is achieved mathematically, based on the assumption
that color space is euclidean. The correlation between visual and calculated
data has been improved by adding empirical functions. Because of the lack of
visual small color difference component data, the degree to which calculated
components are in agreement with visual data is unknown. This is an area that
requires attention.
Surprisingly it appears that the assumption that perceived, properly scaled
 PERCEPTION OF COLOR DIFFERENCES 155

attribute differences add up by euclidean summation has never been explic-

itly tested and confirmed. One problem is that, as seen earlier, there are no
reliable, replicated scales of perceptual hue, chroma, and lightness, a prereq-
uisite for such a test.
Another effort in developing a color difference formula based on visually
determined components has been made by Indow (1999a, 2002). Using visual
judgments of Munsell differences determined for the purpose of multidimen-
sional scaling analysis (as described in Indow, 1988), he attempted to find the
form of geometrical space that would result in the best agreement between
interpoint distances of colors in that space and of those visually determined.
Indow found that the formula he developed resulted in smaller root mean
square (RMS) error than that of Adams-Nickerson but larger than the CIE94
formula. This is surprising because the differences judged were rather large,
involving Munsell chips. In the same year Indow (1999c) also investigated the
accuracy of prediction of Munsell type color differences from the judged
differences in their principal hue components and lightness differences.
Indow obtained the lowest RMS error with a formula that is the sum of
weighted value difference and individually weighted principal hue component
differences:

d˜ (DV , Dx ) = aV DV + (d0 + Saa Dxa ), (4-3)

where d̃ is the predictor of the color difference, DV is the Munsell value

difference, Dxa is the difference in a principal hue component, and aa is the
corresponding weight for the principal component difference, with the values
aR = 0.199, aY = 0.031, aG = 0.098, aB = 0.136, d0 = 0.610, and aV = 0.459. The
principal hue components are functions of value and chroma.
Also in Indow’s work, with different data and using principal hue compo-
nents, simple addition of the component differences was found to give best
results. Because of the normalization of perceived principal hue components
to a maximum value of ten different weighting of the individual components
is required. In differences the red principal component has the strongest effect.
It takes 1.5 units of blue principal hue component (PHC) to equal in a dif-
ference one unit red PHC, 2.2 of green PHC, and, adjusted for the lower
absolute yellow maximum, 3.4 units yellow PHC.
Indow’s work with principal hue components implies, unsurprisingly, that
they are the psychological fundamentals constituting hue. At the same time
they implicitly constitute an aspect of chroma. The achromatic difference to
the maximum value of 10 is also a conflation, in this case of an achromatic
portion of chroma and lightness differences. The lightness difference (pre-
sumably including the Helmholtz-Kohlrausch effect) is, in addition, separately
judged. As mentioned above, Indow’s observers, however, had difficulties dis-
tinguishing in terms of principal hue components between different Munsell
chroma levels at the same lightness.
In this chapter it was demonstrated that for purposes of an object color
156 COLOR ATTRIBUTES AND PERCEPTUAL ATTRIBUTE SCALING

space uniform in color differences, hue, lightness, and chroma appear to be the
essential color attributes. Other attributes can replace lightness and chroma if
uniformity in terms of difference is not the guiding principle. Historically two
sets of fundamental attributes of color perceptions have been proposed. Hue
is common to both. In Hering’s approach veiling is described by the relative
amounts of blackness and whiteness in the perception. But veiling can also be
described in terms of lightness and chroma. It has not been determined how
under comparable conditions the two sets of veiling attributes compare.
Various experimental scales of constant chroma and equal-sized hue differ-
ences vary considerably for unclear reasons. It has also been shown that a psy-
chological color space with the unique hues on the chromatic axes is not
uniform because there are varying numbers of equal-sized hue differences in
the four quadrants. The relationship between chroma and hue differences
appears to change significantly as a function of the size of the differences. In
the next chapter these and other issues are investigated in terms of quantita-
tive stimulus differences.
 Chapter 5
Psychophysical Scaling of
Color Attributes: Stimulus
and Perception

In Chapter 4 color scaling was considered from a purely psychological point

of view. Once attributes were selected, judgments were made using samples
of colored materials to exemplify scales of the attributes in conformance
with the geometrical form selected for the color solid. In a similar manner it
was also possible to study thresholds and suprathreshold color differences
without resort to physical measurements. It was of interest to the early psy-
chophysicists to find the relationship between stimulus magnitude and
response. It has been indicated before that such relationships can only be valid
for tightly circumscribed conditions of viewing and that they can be signifi-
cantly different for other conditions. Within these limits it is desirable to
characterize both stimulus and resulting perception with a high level of
accuracy and repeatability. Since the characterization of the stimulus involves
physical measurements, progress was linked to progress in photometry and
spectrophotometry.

5.1 REQUIREMENTS FOR A UNIFORM PSYCHOPHYSICAL

COLOR SPACE

A formula representing a uniform psychophysical color space for trichromats

must transform spectral power distributions or reflectances in such a way that

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

157
158 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

color scales in three dimensions in good agreement with psychological scales

are created. Not only must the colors, for example of a color circle, be placed
in the correct ordinal order, but they must also be correctly ordered in regard
to saturation or chroma, brightness or lightness, and the unit differences in
hue, chroma and lightness, must be the same. As we will see, simple psy-
chophysical spaces such as a form of cone response space, the CIE X, Y, Z
space, or linearly related opponent color spaces place colors in proper ordinal
sequence; however, distances in these spaces are not proportional to psycho-
logical distances. Since the beginning of the twentieth century a search has
been on to find the transformation that might accomplish this task. It has
become evident that a single formula cannot accomplish it for any kind of
color difference, or only perhaps with several adjustable variables. We begin
by separately considering the three conventional key attributes of psycholog-
ical object color order: lightness, hue, and chroma.

5.2 POSTULATED RELATIONSHIP BETWEEN PSYCHOLOGICAL

AND PHYSICAL MAGNITUDES

Historically postulated relationships between psychological and physical

magnitudes are fraught with difficulties. These are due to the absence of a val-
idated mechanism connecting the psychological experiences of color with the
physical measurement of spectral power. The problems begin with the defini-
tion of the brightness function. Additive results (i.e., where two stimuli mixed
together are seen as resulting in a brightness perception that is the sum of the
perceptions of the two component lights) are only obtained if the spectral
brightness function is determined according to a particular methodology (see
below). Brightness and lightness perceptions thus are complex and dependent
on experimental conditions.
The CIE colorimetric system, developed in the late 1920s, is based on exper-
iments establishing what lights are matching. It does not have an explicit con-
nection to the appearance of stimuli. Such a connection was usually assumed,
however (e.g., see Adams, 1942). The relative success of this assumption has
strengthened the belief in its applicability. Today models of color vision are
usually expressed in terms of cone sensitivity functions. These are believed to
be linearly related to color-matching functions, and vice versa. Models of color
vision have become more sophisticated by considering surrounds and other
experimental conditions, but the fundamental assumption of the direct rela-
tionship between color matching and color appearance remains.
Brightness and lightness and their differences are expressed in terms of
luminance or luminous reflectance, even though this does not consider the
spectral brightness component of the Helmholtz-Kohlrausch effect (see
below). Only the OSA-UCS system has included this effect. Conventionally
hue and chroma are believed to be functions of properly scaled opponent
signals derived from CIE tristimulus values, hue the ratio and chroma the
euclidean sum (see Section 5.6). These are empirical approximations with a
 PHOTOMETRY AND BRIGHTNESS/LIGHTNESS 159

degree of validation from the correlation with visual data, as we will see. To
bring such models into close agreement with psychological results requires
extensive fine-tuning and application of auxiliary functions, as Chapter 6 will
show. The following sections discuss the development of these implied rela-
tionships in some detail.

5.3 PHOTOMETRY AND BRIGHTNESS/LIGHTNESS

For the visual sense the earliest quantitative studies comparing stimulus to
response involved brightness. Photometry, the quantitative measurement of
light, has been of interest since the Renaissance. In its more modern form it
was developed by Bouguer and advanced by Lambert and Benjamin Thomp-
son, the Count of Rumford (1753–1814). The initial equipment consisted of
simple wedges that allowed visual comparison of two different light sources
under controlled conditions. A curious, very early illustration of what appears
to be a photometric measurement1 is that by the celebrated painter Rubens
used as the frontispiece for Book V of d’Aguilon’s text of 1613 Opticorum
Libri Sex (Fig. 5-1). Here the effect of distance of light source on the perceived
brightness of the patches on the screen is being studied.
Bouguer made quantitative measurements by comparing a test light source
against a candle as the reference source. The first step toward making mea-
surements more reliable was to use an artificial reference light source other
than a candle. This resulted, beginning in the late eighteenth century, in the

Fig. 5-1 Rubens’s, frontispiece illustration to Book V of d’Aguilon’s Opticorum libri sex depict-
ing an experiment in measuring light intensity as a function of distance of the lights, 1613.
160 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

development of oil lamps, pentane lamps, and the Hefner lamp, burning
various hydrocarbons (used until the 1940s).2 The human observer began to
be replaced by photoelectric cells beginning in the early twentieth century.
Lambert had already envisaged this development by describing in Photome-
tria in concept a piece of equipment that could measure light analogous to
temperature measurement with a thermometer.3 For such measurements to be
in agreement with visual measurements the correspondence between the spec-
tral sensitivity of photoelectric cells and that of the average human had to be
established.
The discovery of rods and cones of the human visual system in 1828 by the
German physiologist G. R. Treviranus (1776–1837) and the discovery of visual
purple in 1877 by F. Boll resulted in extended efforts to determine their
purpose and activity.4 The duplicity theory of the double function of the retina,
according to which the cones mediate daylight vision and the rods night vision,
was proposed by the German anatomist M. J. S. Schultze (1825–1874) in 1866
and expanded by Helmholtz’s student Kries. Measurements of the luminosity
of spectral colors at different levels of illumination were made in Helmholtz’s
laboratory by König and C. Dieterici (1884, 1892). König was able to show
the close agreement among an absorption curve he had measured for visual
purple, the luminosity curve of a person with monochromatic vision estab-
lished under daylight conditions, and the luminosity curve for an observer with
normal vision determined at very low light levels. This was a clear indication
of the existence of the postulated second visual system based on rods and
operational for night vision in observers with normal color vision. König and
Dieterici obtained a different spectral sensitivity curve for daylight vision.
Comparison of the sensitivity of photoelectric cells and of humans indicated
large discrepancies, and in order to bring measurements into agreement with
average visual results, filters had to be interposed between light source and
the photoelectric cell.
A new type of photometer, the flicker photometer, appeared in the early
twentieth century and was intensively investigated by H. E. Ives (1912). The
original method of obtaining flicker was to rotate a partial disk in front of a
light. Depending on the disk sector and the speed of the disk, more or less
flicker is observed. If the occluding portion is half of the disk, at the critical
flicker frequency at which the flicker disappears the perceived intensity of light
is exactly half the intensity when viewing it unobstructed. By matching the
brightness of chromatic flickering fields against a steady field of white light,
the luminous efficiency of chromatic lights can be measured. Ives’s studies
showed that brightness measured in this manner is additive. Additivity is not
observed when brightness of lights of different color is measured by compar-
ing steady fields. Given the state of computing at the time, additive functions
were very desirable. Important determinations of spectral brightness using
flicker photometry were reported in 1923 by K. S. Gibson and E. P. T. Tyndall,
working at the U.S. National Bureau of Standards. An average sensitivity curve
was recommended by them to the CIE and standardized largely unchanged
 THE COLORIMETRIC SYSTEM 161

Fig. 5-2 CIE standard scotopic (low light, V¢l) and photopic (daylight level, Vl) luminosity
curves. The wavelengths of maximum sensitivity are indicated. From Wyszecki and Stiles
(1982).

by that organization in 1924 as the CIE standard spectral luminous efficiency

function V(l) (CIE, 1924), representing the average photopic observer (day-
light vision). A comparable scotopic standard observer (low-light vision) was
defined in 1951.The spectral sensitivity functions of the two observers are illus-
trated in Fig. 5-2. The CIE luminosity function has been criticized after its stan-
dardization (e.g., see Stockman and Sharpe, 1999), and in 1988 the CIE offered
a revised version based on proposals by Judd in 1951 and by J. J. Vos in 1978
(CIE, 1990). The 1924 function nevertheless is an integral part of the CIE
colorimetric system and continues to be widely used.

5.4 THE COLORIMETRIC SYSTEM

The idea of trichromatic color vision was originally put forth by Palmer in 1777
and revived by Young in 1802. Important support was later provided by the
work of Helmholtz, Grassmann, and Maxwell, as discussed earlier. In the 1850s
Helmholtz estimated the spectral sensitivity of the human color vision system
in agreement with Young’s theory (Fig. 5-3; Helmholtz, 1860). In 1860 Maxwell
provided the first set of measured functions derived using a visual colorime-
ter (Fig. 5-4). Detailed measurements were made by König and Dieterici in
1892, using an advanced König-Helmholtz spectral colorimeter (Fig. 5-5).
Similar measurements were made also by Abney and reported in 1914.
162 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-3 Helmholtz’s sketch of estimated spectral sensitivity of three fundamental color vision
processes, 1860. The letters refer to major hues of the spectrum, from red ( R) on the left to
violet on the right.

Fig. 5-4 Spectral curves of the three fundamental processes of color vision R, G, and B, and
the sum curve S as determined by Maxwell, observer K. The wavelength scale is in arbitrary
units; C–G identify Fraunhofer lines in the spectrum, 1860.

Between 1920 and 1930 J. Guild (1930) and W. D. Wright (1928–29) in England
independently built improved equipment to measure these functions, and the
color-matching functions of various observers were determined. These data
were considered by the CIE for standardization. Following a suggestion by
Judd, linear transformations were calculated so that two of the implied pri-
maries were located on a line with zero luminance (the alychne). This resulted
in the third primary having a spectral function identical to the CIE luminance
function. In this form the functions were standardized as the CIE 1931 2° stan-
dard observer color-matching functions (applicable to a field of view of 2°, Fig.
5-6; CIE, 1931).5 In 1964 the CIE 1964 10° standard observer color-matching
functions were added, applying to a field of view of 10° (CIE, 1964) and based
 THE COLORIMETRIC SYSTEM 163

Fig. 5-5 Measurements of their own fundamental color vision processes R, G, and V (violet)
by König and Dieterici. The dashed line represents the G function of an observer with anom-
alous color vision, 1886. Note that the wavelength scale is in reverse. Letters along the wave-
length scale denote major Fraunhofer lines.

Fig. 5-6 CIE spectral color matching functions of the 2° standard observer. They have been
linearly transformed from the measured functions. From Wyszecki and Stiles (1982).

on measurements by W. S. Stiles and J. M. Burch (1955) and by N. I. Speran-

skaya (1959). The two sets of functions differ somewhat.6 With these defini-
tions of standard observers, as well as definitions of illuminants and of the
reflectance factor defining the reflecting properties of object colors, a system
was available for quantitative colorimetry. In this system any colored mater-
ial as viewed under a standard light source by a standard observer is expressed
with three figures identified as the tristimulus values X, Y, and Z, where Y is
164 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

identical to the luminous reflectance for object colors (luminance for lights).
The X, Y, Z space is a nonuniform psychophysical color space, taken to be
euclidean, relating measured reflectance of a sample viewed in a standard light
to a specific point in the space. Tristimulus values in case of object colors are
normalized sums of the reflectance function weighted by the spectral power
distribution of the light source and the three color matching functions x̄, ȳ, z̄.
With one dimension in this space representing (flicker) luminance, the other
two must represent in some fashion the chromatic components from which
hue and chroma are derived. In the CIE colorimetric system the psychophys-
ical definition of hue is the dominant, or complementary wavelength, of satu-
ration colorimetric purity as expressed in the CIE chromaticity diagram, a
particular version of chromatic plane defined by the chromaticity coordinates
x and y. Their definition is as follows:

X Y (5-1)
x= and y= .
X +Y + Z X +Y + Z

In the chromaticity diagram spectral colors fall on a horseshoe-shaped func-

tion, with saturated nonspectral purple colors falling on a straight line con-
necting the ends of the spectral function (Fig. 5-7). All possible object and light
colors fall on or within the outline of the diagram. The chromaticity diagram
is also not visually uniform. A color space can be constructed with the axes x,
y, Y. It is commonly used to represent color relationships. An example is the
Rösch-MacAdam solid illustrating on its surface optimal object colors for a
given standard observer/illuminant combination (see Fig. 2-46). The X, Y, Z
space and the chromaticity diagram allow the determination of quantitative
relationships between colors in a linear system, but without consideration
of surrounds. The locus of spectral colors in X, Y, Z space is illustrated in
Fig. 5-8.
The CIE colorimetric system has not escaped criticism and has been even
ridiculed.7 Despite its weaknesses it remains a central aspect of color science
and its technical applications. Among the reasons are its noticeable technical
successes, the lack of better proposals, and resistance against change. True
progress, if possible, requires knowledge of the contents of the black box con-
necting physical stimuli with the experience of color.

5.5 CONE RESPONSE SPACE

Physiology has identified three cone types in the human retina, in agreement
with the postulates of trichromatic vision (but recall the comment about
females with the genetic potential for four cone types in Chapter 1). These
three cone types are presumably the only transducers of information respon-
sible for color vision (rods may contribute to color vision under certain limited
circumstances). Along the visual path in the brain many transformations of
 CONE RESPONSE SPACE 165

Fig. 5-7 CIE chromaticity diagram with spectral trace and the nonspectral purple colors for
the 2° standard observer. E represents the equal energy illuminant. From Wyszecki and Stiles
(1982).

the information generated in the retina take place, all derived from cone
responses. Cone sensitivity and cone response to specific light energy are
therefore important pieces of information in an attempt to link physical stim-
ulus and psychological response. Attempts to measure spectral absorption
functions of cone pigments began in the middle of the twentieth century, with
refinement continuing into the present. However, light radiation undergoes
changes in the eye before being absorbed by the cones of the retina, changes
that need to be accounted for. They are due to the media in the eye from the
outer surface to the cones. For a comparison between stimulus and percep-
tion, cone response functions taking into account intraocular absorptions are
therefore essential. In vivo measurements of human cone response are not
practical, and the data currently accepted are based on absorption measure-
ments of excised eyes, on measurements made on primates, and on the as-
sumption that cone response functions are linearly related to color-matching
functions. Proposals for standard cone sensitivity functions have been made
166 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-8 Locus of spectral colors in the CIE X, Y, Z space, beginning at 380 nm and ending
at 750 nm (wavelengths 500 and 600 nm are identified).

by V. C. Smith and J. Pokorny (1975), by Vos (1978), by A. Stockman et al.

(1993), and by Stockman and L. T. Sharpe (1998). The latest proposals are
based on the Stiles-Burch color-matching functions. There are relatively small
differences between the three sets of data that for the current purposes do not
seem to be of significance, however, even though, as discussed below, they
could have a noticeable effect on correlation with hue difference perception.
Cone response functions are reported in different ways: linear or logarithmic
scales, function peaks at unity, or identical area under the curves. Figure 5-9a
and b illustrate the Smith-Pokorny functions (as derived from the CIE 2°
observer data) in linear scale as well as in logarithmic scale. In the past
the three functions were often called R, G, and B, for red, green, and blue.
In recent years, as more details about the opponent color system became
known, it has been realized that this is not an appropriate way of identi-
fying them. They are now called L, M, and S (for long, medium, and short
waveband sensitivity). The three functions can be thought of as constituting
the axes of a (nonuniform) psychophysical color space. Figure 5-10 illustrates
the spectral trace in the orthogonal L, M, S space defined by the Smith and
Pokorny functions. None of the axes of this space has a direct relationship
to a color attribute. An unresolved issue is the effect of field size on the cone
sensitivity functions. If, as said above, cone sensitivity functions are linearly
related to color-matching functions, the cone sensitivity functions must be
different for the 2° and 10° standard observers. There are reasons why this
could be so. Some vision scientists believe that the same transformation
 CONE RESPONSE SPACE 167

Fig. 5-9a Spectral cone sensitivity functions according to Smith and Pokorny, calculated for
the CIE 2° standard observer, with identical area under the curves, linear sensitivity scale.

Fig. 5-9b Spectral cone sensitivities from Fig. 5-9 a in logarithmic sensitivity scale.
168 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-10 Spectral trace in the L, M, S space based on the cone sensitivity functions of Fig.
5-9a. The wavelengths 500 and 600 nm are identified.

formulas are applicable for the 2 and 10 standard observers (Smith,

1999).
Cone responses are essentially in linear relationship with the absorbed light
energy up to a luminance of approximately 1000 cd/m2 (a cloudy sky at noon).
Above that, for example, in bright sunlight, saturation effects occur resulting
in relatively reduced response.

Individual Variation in Cone Functions and Smoothness of Functions

In recent years a significant effort is under way to develop a new set of
cone functions taking into account the latest information on genetic variabil-
ity of cone expression and using modern technological means (Stockman and
Sharpe, 1998, 1999).8 Genetic research has indicated that the L and M cones
can be expressed in different ways and that some people have a hybrid L/M
cone (Sharpe et al., 1999). Wavelengths of peak response for L cones vary in
individuals from approximately 554 to 563 nm.There is less variation in M cone
expression, with peak wavelengths varying from approximately 532 to 538 nm,
and only one normal expression of the S cone. There are differences between
the sexes in the prevalence of genetic expressions of cones. As mentioned in
Chapter 1, it is known that up to 50% of females have the genetic potential
 OPPONENT COLOR SPACE 169

for four cone types. So far no individual actually having four cone types has
been identified. All these issues touch on the question of a statistically mean-
ingful average observer and how different sets of data may be influenced by
the genetic variability in its observer pool. Such observer pools are likely
to also differ in the degree of ocular yellowing and distribution of macular
pigment, both affecting in particular the S cone sensitivity function. Another
issue in connection with cone functions is the question of their intrinsic
smoothness. CIE color-matching functions are smooth, and cone functions
calculated from these by linear transformation are also smooth. Directly
measured individual cone sensitivity functions differ somewhat in shape and
appear to have in some cases more or less pronounced “dents” (see Fig. 5-5
for an example). Some of these appear to be due to the degree of macular
absorption in individuals, as already König surmised.

5.6 OPPONENT COLOR SPACE

Another version of a psychophysical space is the opponent color space. It

attempts to provide a psychophysical model for Hering’s opponent color
hypothesis of color vision and to describe the zone following the cone absorp-
tion level in a multiple-zone model. With an opponent color space the con-
ceptual leap is made that there is a connection between functions predicting
which color stimuli produce identical color experiences (color-matching func-
tions) and color appearance. Such a connection is not a priori a certainty. It
will be shown that such a space can be a rough euclidean approximation of a
uniform color space. However, aside from the issue of euclidean nature there
are other discrepancies of detail arguing for a considerably more complex
picture of reality than a simple opponent color space can provide.
As mentioned before, according to Hering’s hypothesis there are three
opponent processes that are responsible for our color experiences: a
greenness-redness, a yellowness-blueness, and a whiteness-blackness process.
Hering had developed physiological ideas of how such a process could be
implemented. He engaged in an epic struggle with Helmholtz pitting the oppo-
nent color theory against what is known as the Young-Helmholtz trichromatic
theory. After conceptual proposals by Donders and Kries, Helmholtz showed
in 1896 how the two zone components could be expressed in mathematically
equivalent form, but voiced doubts about the need for a second zone step. The
psychologist and Hering supporter G. E. Müller developed a complex three-
stage theory in 1896 to be championed in the 1960s by Judd. Schrödinger
(1887–1961) showed in 1926 again, using the König and Dieterici
spectral sensitivity data, how the trichromatic and the opponent color theory
can be combined. From the sensitivity data he calculated spectral opponent
color response functions (Fig. 5-11). In 1927 Luther in Germany and, inde-
pendently, in 1929 Nyberg in Russia proposed what came to be known as the
Luther-Nyberg color solid (see Section 2.22, Fig. 2-47). It is based on a (nonuni-
170 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-11 Schrödinger’s opponent color and brightness functions calculated from König and
Dieterici’s fundamental functions (Fig. 5-5). Note that the wavelength is reversed from the cus-
tomary presentation, 1926. Hell refers to the brightness function. The other two functions rep-
resent yellowness-blueness ( Gelb, Blau) and greenness-redness ( Grün, Rot).

form) linear opponent color space with two color moments and a color
weight.
In 1923 the American physicist E. Q. Adams proposed a zone theory in
which the output from three cone types is modulated and subtracted to result
in opponent color signals. Adams expanded on his proposal in 1942 where he
showed, using CIE tristimulus data, that a theory such as he had proposed
could offer a good model for the Munsell system (Fig. 5-12). There he specif-
ically proposed that the output of the three cone types should be considered
identical to the CIE tristimulus values and the modulation of the output iden-
tical to that of the Munsell value function. He constructed a “chromance”
diagram by subtracting modulated Y values from modulated X and Z values.
In 1944 Nickerson and K. Stultz offered a color space and color difference
formula based on Adams’s proposals which became influential in technology
(more details in Chapter 6).
At the same time support for an opponent color step in color vision began
to develop on a different front. Around 1940 physiologists began to measure
signals by inserting microelectrodes into retinal cells. The Swedish physiolo-
gist R. Granit measured spectral response curves in ganglion cells in the retina
 OPPONENT COLOR SPACE 171

Fig. 5-12 Munsell colors at value 5 of the 1929 Book of Color in Adams’s 1942 opponent color
chromatic diagram. Lines represent the smoothing proposal by Newhall (1940).

of frogs. His student G. Svaetichin found chromatic opponent cells in fish

retinas in 1956.9 In the 1960s American researchers made measurements from
cells of the lateral geniculate nucleus of monkeys and found various kinds of
cells with opponent color sensitivity (De Valois, 1965; Wiesel and Hubel, 1966).
On the psychological front the American psychologists D. Jameson and L.
Hurvich began in the mid-1950s to use hue cancellation as an experimental
method and deduced opponent chromatic response functions from the results.
In such experiments, for example, a “blue” stimulus is added to a yellowish
color until the perceived yellowness is canceled or neutralized. They were able
to show that their personal experimental functions could be reasonably
approximated using CIE standard observer color-matching functions (Fig. 5-
13) in a manner employed by Adams and by Judd in his interpretation of the
Müller theory (Hurvich and Jameson, 1955; see Chapter 6 for more details).
In 1964 Hurvich and Jameson also translated into English Hering’s posthu-
mously published Grundzüge der Lehre vom Lichtsinn (Outlines of a theory
of the light sense; Hering 1905–1911).
If hue changes depend directly on the chromatic responses shown in
Fig. 5-13, it is apparent that they vary considerably as a function of wavelength.
172 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-13 Opponent color functions of the CIE 2° standard observer calculated by Jameson
and Hurvich (1955).

Near 475 nm the greenness-redness signal is at zero, indicating the presumed

location of unique blue for the CIE standard observer. The blueness signal on
both sides of that point changes rapidly, as does the greenness-redness signal,
indicating rapid hue changes per nanometer increment. This is indeed the case
as wavelength discrimination experiments show. At approximately 500 nm is
the neutral point of the blueness-yellowness signal (unique green, but see
Section 5.10). Also here both signals change relatively rapidly indicating con-
tinued high sensitivity to wavelength changes. By this conjecture the complete
change from unique blue to unique green takes place within about 25 nm. At
about 578 nm the greenness-redness signal passes through its second neutral
point resulting in the display of unique yellow. While the yellowness signal
changes relatively slowly in this region, the greenness-redness signal changes
rapidly on both sides of the neutral point indicating rapid changes toward
greenish yellow and reddish yellow away from the neutral point. This is again
supported by a high rate of change in wavelength discrimination experiments.
In the regions of approximately 440, 530, and 610 nm and above the rate of
change is significantly lower, resulting in reduced perceived wavelength dis-
crimination. While a reasonable agreement exists between the form of oppo-
nent functions and experimental wavelength discrimination, there is no solid
agreement between rate of change data behind these functions and experi-
mental hue difference data, as will be shown below. There is in addition the
matter of the wide ranges of wavelengths of the blue, green, and yellow unique
hues as determined by individual observers.
 OPPONENT COLOR SPACE 173

Fig. 5-14 Spectral trace in the Y, a*, b* space, 10° observer. The wavelengths of 500 and
600 nm are identified.

As Chapter 6 will show, most color vision models developed since the mid-
nineteenth century are zone models and a simple version of a zone model
color space was recommended in 1976 by the CIE as the CIELAB formula
(CIE 1976; see Chapter 6). In this space, considered euclidean and with a carte-
sian chromatic diagram, colors are identified by chromatic coordinates a* and
b* and a lightness coordinate L*. There is also a polar coordinate interpreta-
tion where chroma is defined as the radial distance from the origin and hue
by the hue angle, thus bringing the system conceptually in agreement with a
psychological system of the Munsell type. Experimental work before and since
then has indicated that there are surround effects requiring consideration and
that CIELAB space is at best a rough approximation of a large difference
uniform color space. Figure 5-14 illustrates the spectral colors in the Y, a*, b*
space.
An important matter usually not considered in models until very recently
is that of the interaction of cone responses between test field(s) and surround.
Four types of opponent cells involving L and M cone output comparison have
been identified (Wässle et al., 1994). All four involve comparison of output of
cones from the center of the receptive field of the opponent cell with those
from the surround. A similar situation applies to the opponent cells compar-
174 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

ing output from S cones with that of L plus M cones.These mechanisms appear
to be responsible for test field/surround effects, such as contrast and assimila-
tion, possibly crispening, and others. Correct treatment of these effects is
essential for a uniform color space model.
In recent years experimental data have begun to raise doubts about a
simple subtractive opponent color system such as proposed by Hurvich and
Jameson. New experimental work points to four independent chromatic
dimensions and perhaps to a multiplicity of hue detection mechanisms (see
below).

5.7 HOW ARE THE L, M, S AND X, Y, Z COLOR SPACES RELATED?

On a mathematical basis, according to Smith and Pokorny (1975), the two

spaces are related by a set of linear transformation equations as follows:

l = ( 0.15516 x + 0.54308 y - 0.03287z ) ,

m = ( -0.15516 x + 0.45692y + 0.03287z ) ,

s = 1.00000z , (5-2)

where x̄, ȳ, z̄ are CIE color-matching functions, originally the Judd and Vos
modified 2° observer functions. In this transformation l̄ + m–– add up to equal
ȳ (equivalent of the luminosity function). As mentioned, the same equations
have been applied to both CIE 2° and 10° observer data. Slightly different
equations have been proposed by other authors.
The inverse relationship is as follows:

x = 2.920l - 3.445m + 0.208 s ,

y = l + m,

z = 1.000 s . (5-3)

Both the L, M, S space and the CIE tristimulus space are viewed as orthogo-
nal. It is evident that the two spaces cannot be orthogonal in the same refer-
ence frame, but a case can be made that they represent different realities and
each may be orthogonal by itself. There is, however, the unanswered basic
question if the assumption of orthogonality is valid for either space.
Based on their recordings from parvocellular neurons in the LGN of
macaques A. M. Derrington, J. Krauskopf, and P. Lennie (DKL) in 1984 pro-
posed a color space (nonuniform) representative of the signal output of dif-
ferent cell types (see Chapter 6). They located opponent cells that report in
terms of L–M or the reverse and in terms of (L + M) - S or the reverse. When
they plotted the corresponding axes in the CIE chromaticity diagram, they
 HOW ARE THE L, M, S, AND X, Y, Z, COLOR SPACES RELATED? 175

Fig. 5-15 Balanced opponent functions a and b calculated from CIE color-matching functions.
These functions are supported by recordings in the lateral geniculate nuclei of macaques.

found that the axes do not represent average unique hues. This can be demon-
strated by calculating balanced (equal areas under the four curve segments)
opponent functions based on cell input reported by DKL:

a = 5.673(l - 1.98842 m ),

b = l + m - s. (5-4)

The resulting two functions a and b are illustrated in Fig. 5-15. The b function
is identical to ȳ - z̄ and therefore has the same form as the Hurvich-Jameson
yellowness-blueness opponent color function. The a function is different from
the greenness-redness function because it lacks positive values in the short-
wave range, considered indicative of the reappearance of redness at short
wavelengths. The a function can therefore not be a greenness-redness express-
ing function (in the sense that green and red are represented by the unique
hues), and the DKL space is not a space in alignment with the psychological
space (e.g., see Lee 1999). It is important to be aware that the measured func-
tions apply to macaques and that they are from the lateral geniculate nucleus:
details in the human visual system may differ and additional transformations
in the cortex are known to take place. Recently it has been found that in the
primary visual cortex at the back of the brain, most cells receive input from
all the different opponent cell types in the lateral geniculate nuclei (De Valois
176 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

et al., 1997). On that basis a mechanism for S cone input into the greenness-
redness opponent system appears to exist. Equation (5-3) indicates that the
definition in terms of cone sensitivities of the x̄ color-matching function
includes S cone input. It is interesting to note that axis rotation required to
provide a result equal to a balanced linear opponent chromatic diagram (based
on CIE 1931 tristimulus values) can be achieved by the following equation
(Kuehni, 2000a):

a = f (a - 0.56b ),

b = b, (5-5)

where a and b are linear opponent color functions, f is a normalization con-

stant with the value 0.989. The factor for b is close to 0.5. It will be shown later
that this factor varies considerably for optimal modeling of various sets of
visual data. The subtractive step is reminiscent of the second transformation
in the G. E. Müller theory (see Chapter 6).

5.8 EXPRESSING PSYCHOLOGICAL SCALES IN

PSYCHOPHYSICAL SPACES

Historically the X, Y, Z space (and its predecessors) has been used as a basis
to express psychological scales in a psychophysical space. The fact that one of
its dimensions is aligned with brightness and the other two therefore with
chromaticness seemed a good reason to expect reasonable agreement if the
functions expressing x̄ and z̄ are correctly selected. They were selected so that
they had only positive values (the experimentally determined color-matching
functions have stretches of negative values). There is, of course, no a priori
reason why a color-matching space should be in agreement with, or easily
modifiable into, a color appearance space, and as will be seen later, there are
important discrepancies.

Lightness Scales Expressed in Terms of Luminous Reflectance

Seemingly the first lightness scale that through reconstruction can be
expressed in terms of luminous reflectance is Glisson’s gray scale. Figure 2-9
shows the progression of CIELAB L* lightness values (derived from
luminous reflectance) as a function of the grades of the reconstructed scale.
Insufficient quantitative data are available to deduce lightness scales in
terms of luminous reflectance for the Mayer, Lambert, Runge, or Chevreul
systems. As discussed in Chapter 4, Plateau in 1872 concluded that ratios were
applicable, based in part on the apparent preservation of the contrasts in etch-
ings in different lights, and that as a result the relationship would be linearized
by a power function. Delboeuf, in an elaborate experiment of halving the
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 177

visual distances between two grays independently, concluded that a logarith-

mic formula provided the best fit to his data. Plateau received confirmation in
1887 from experiments by Breton who found a square root power relation-
ship to be applicable. But, as we have seen, the Weber-Fechner law was the
generally accepted paradigm for some 100 years.
Ebbinghaus’s eight-grade gray scale had the following relative luminance
values: 100, 44.4, 21.1, 10.3, 5.8, 4.4, 2.0, 1.0 (Ebbinghaus, 1902). This series is
linearized optimally with a 0.02 power. The resulting scale is not quite uniform
and does not quite represent a Fechner type log scale. Ridgeway’s gray scale
incorporated into his 1912 color atlas had, by definition, a logarithmic step
relationship as did Ostwald’s gray scale of 1917.
Priest, Gibson, and McNicholas investigated the relationship between
Munsell gray scale steps in the 1915 atlas and luminous reflectance values
(Priest et al., 1920). They found that the relationship is best described by a
square root power function. Taking the 1919 National Bureau of Standards
measurements of the neutral samples of the 1915 Atlas of the Munsell Color
System, as well as the 1926 diffuse measurements of the same samples (Gibson
and Nickerson, 1940; see Fig. 7-5), the optimal linearization (minimal coeffi-
cient of variation) is found to be based on an 0.42 power. However, the func-
tion plotting reflectance against value step is not smooth. It is composed of
two segments intersecting between value steps 4 and 5. The lower segment is
optimally linearized with a power of 0.25, the upper with one of 0.48. This
appears to be an indication of lightness crispening (see below), as one would
expect, and points to a surround with a luminous reflectance of approximately
Y = 20.
In 1922 Adams and his colleague P. W. Cobb studied the effect of surround
on perceived brightness and developed a formula (in the version reformulated
by Judd) that recognizes the effect of the luminous reflectance of the surround
on the results:

Ê Rb + 100 ˆ
V =R , (5-6)
Ë R + Rb ¯

where V is lightness on a scale of 0 to 100; R is luminous reflectance in per-

centage of the sample, and Rb luminous reflectance of the surround.
A. E. O. Munsell, L. L. Sloan, and I. H. Godlove continued work on the
Munsell value scale and described in 1933 results based on a combination of
“just noticeable differences” as well as scale-halving experiments. Their scale
is similar to the Munsell Atlas and 1929 Book of Color scales. It is optimally
linearized by a 0.34 power but consists of two branches with the lower lin-
earized with an 0.50 power and the upper with an 0.20 power. Godlove (1933)
compared it to the Adams-Cobb data and fit the following empirical formula
to the Munsell data, a formula described as “intermediate between those of
the Fechner-Delboeuf [logarithmic] and Plateau-Munsell [power function]
equations”:
178 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-16 Results of visual determination of Munsell value of Munsell value scale samples
viewed against a white, mid-level gray, and black background. After Newhall et al. (1943).

V 2 = mR + nR 2 , (5-7)
where V is the Munsell value, m and n are constants, and R is luminous
reflectance.
In the studies leading up to the Munsell Renotations the value scale was
evaluated against a black, a middle gray, and a white background. The results
differed significantly (Fig. 5-16). The optimal power for the results against the
black background is cube root against the white background square root. As
expected, the results against the gray background show two branches, indi-
cating the lightness crispening effect. The committee working on the Renota-
tions decided to average the results in some fashion, and the final, smoothed
curve was mathematically described with a quintic function developed by
Judd:

Y = 1.2219V - 0.23111V 2 + 0.2395V 3 - 0.021009V 4 + 0.0008404V 5 , (5-8)

where V is the Munsell value (Newhall et al., 1943). Glasser and colleagues
showed in 1958 that this function is closely matched by a cube root function.
Figure 5-17 illustrates lightness scales based on logarithmic (Weber-
Fechner) and cube root functions as well as an Adams-Cobb function with a
surround luminous reflectance of 35%. It indicates that power functions have
less modulating effect than the Weber-Fechner function and that an Adams-
Cobb function with a surround luminous reflectance of 35% is roughly com-
parable to a cube root function.
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 179

Fig. 5-17 Luminous reflectance Y plotted as a function of lightness according to logarithmic,

cube root, square root scaling, as well as the Adams-Cobb lightness scale for a surround of
luminous reflectance Y = 35.

As mentioned before, in 1933 Schönfelder investigated the effect of sur-

round variation on the precision of color matching. He found that precision is
highest if the surround color is intermediate to the standard and the match
fields. This applies not only to lightness but also to chromatic differences,
and we will encounter what Judd called Schönfelder’s law again later.
Schönfelder’s results were in 1938 confirmed by K. J. W. Craik for achromatic
colors and in 1952 by W. R. J. Brown for chromatic colors.
In 1964 Jameson and Hurvich investigated brightness scaling, and they
introduced a subtractive constant that was explained as correcting for simul-
taneous contrast effects:

V = bx p - V0 , (5-9)
where V0 is the subtractive constant. In the same year T. Kaneko investigated
lightness of achromatic colors in form of color chips as a function of surround
lightness. He found, in agreement with Schönfelder, that differences between
gray scale grades were perceived largest if the surround lightness falls between
the lightnesses of the samples being compared. When the sample lightness is
higher or lower than the surround lightness, a larger increment in luminous
reflectance is required for a perceived difference of the same magnitude. He
also found that the effect depends to some degree on the size of the test fields
compared to the surround. With larger test fields the crispening effect was
reduced; that is to say, the magnitude of the lightness crispening effect appears
180 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

to be a function of the test field size, the effect being strongest for small fields.
Kaneko constructed gray scales valid for a given surround lightness (Kaneko,
1964). Adams and Cobb had also noticed the described effect, but they
believed it to be limited to threshold level differences. Kaneko found it also
to apply for much larger lightness differences. It was H. Takasaki who gave the
effect the name lightness crispening. Kaneko developed a formula consisting
of two simultaneous equations to model his results.
Takasaki conducted in 1966 extensive investigations of the effect of
surround lightness on perceived lightness of test fields. He confirmed contrast
as well as crispening effects. His complex formula was shown by C. C.
Semmelroth in 1970 to be reducible to
n
V = Rm + k R - Rb , (5-10)

where V is the perceived lightness, R is the luminous reflectance of the sample,

and Rb that of the surround, m and n are exponents, and k is a constant. The
equation applies for situations where R is equal or larger than Rb. In the oppo-
site case the +sign is replaced with a -sign. For various experimental data sets
m = 0.4, n = 0.2, and k = 0.65 have been found to provide good fits.
In 1975 Judd and Wyszecki constructed an instructive graph of the effect of
lightness crispening on perception of value differences in the Munsell value
scale, as described by equation (5-10) (Fig. 5-18). Explanatory example: the
lightness difference between two Munsell samples Vs = 2 and Vs = 3 when
viewed against a surround of value 7 is approximately 1.65 - 0.7 = 0.95 units.
When viewing the samples against a surround of value 3, the corresponding
difference is 3.0 - 1.2 = 1.8 units, or nearly twice perceived size. More recently
lightness crispening has been measured and described by Whittle (1992).
The effect of surround on perceived lightness is demonstrated with the
results of an experiment with 22 observers. On a video screen in a dark room,
successively against black, white, and gray (L* = 45) backgrounds, the
observers adjusted the luminances of eight separated fields so that all steps
looked perceptually equidistant. Grade 4 had against the gray surround the
same lightness as the surround itself. For the other two surrounds the lumi-
nance of grade 4 was adjusted so that it had the same lightness appearance as
in the gray surround. The average result in terms of CIELAB L* lightness is
illustrated in Fig. 5-19. The inflection of the crispening effect at the luminance
of the surround is clearly visible. It is apparent that the L* scale is not applic-
able to any surround lightness. Similar results were obtained with chromatic
fields. Lightness crispening has been investigated by Lübbe (1999) and iden-
tified in small color difference data by Kuehni (2001b). In agglomerations of
data from different experiments made with different surrounds, lightness
crispening is obscured. In several individual data sets it is clearly detectable.
It was, for a single surround, measured by Chou and colleagues (2001). For
small color difference formulas, several different weightings of the CIELAB
lightness scale have been proposed since the mid-1970s, as shown in Chapter
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 181

Fig. 5-18 Graph developed by Judd and Wyszecki from a table by Semmelroth (1971)
depicting the effect of lightness crispening on perception of value differences in the Munsell
Renotation value scale. From Judd and Wyszecki (1975). Vs is the nominal Munsell value, Va
is the background adjusted Munsell value.

6. Surprisingly few formulas consider surround lightness explicitly (Lübbe,

1999; Kuehni, 2001d) where it has improved the correlation between visual
and calculated color differences (Kuehni, 2001b).
Some researchers recently began to use the term masking for the crispen-
ing effect. Stimuli producing a strong contrast to the surround are said to
produce masking and require larger increments for detection than those at
lower contrast (Eskew et al., 1999). Aside from luminance or luminous
reflectance masking there is also chromatic masking, discussed below. In such
experiments masking is reported to produce threshold elevations that can be
approximated with a power law with exponents from 0.5 to 1.0. Such masking
experiments typically involve only brief exposures to the stimuli and other
controls that minimize adaptation and other effects. The results therefore
cannot be expected to closely match those of experiments involving object
color samples with unrestricted exposure.
182 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-19 Average settings expressed in L* values of 22 observers of the luminance of eight
grades of achromatic colors against three surrounds so that steps between the grades appear
equidistant.

The CIELAB formula of 1976 enshrined the Munsell value function in a

cube root format to connect luminous reflectance with perceived lightness. The
OSA-UCS formula, as will be seen in Chapter 6, used the Munsell value func-
tion as a basis but with adjustments that consider lightness crispening and the
Helmholtz-Kohlrausch effect, the latter to be discussed presently.
The gray scale of the Swedish Natural Color System was obtained by
assessment of a series of gray samples and their percentages of innate ideas
of the observers of ideal white and black. The Adams-Cobb formula with a
surround luminous reflectance of Y = 56 was found to fit the experimental data
well (Hård et al., 1996).
It should be mentioned here that a parallel process, chromatic crispening,
exists. According to this effect the smallest stimulus increment for a criterion
chromatic difference is required if the chromaticities of the two test samples
straddles the chromaticity of the surround (see Chapter 8).
The work of Gilchrist and colleagues (1999) and of Purves and colleagues
(see Purves and Lotto, 2002) in regard to the complex relationship between
luminance and brightness/lightness was mentioned in Chapter 4. In the
Munsell and similar scales the luminance ratio between white and black is
close to 100 : 1, and according to the Gilchrist group’s findings, luminance scale
compression takes place under these conditions, just as it is found in most data
of that kind.
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 183

Heterochromatic Brightness/Lightness

When luminance measurements of colored stimuli began to be made routinely,

it became evident that the perceptual brightness of constant luminance stimuli
was higher than that of an achromatic color with equal luminance. The chro-
matic stimulus appears to glow compared to the achromatic one. Expressed
conversely for object colors, highly chromatic objects having the same per-
ceived lightness as a gray object have a reduced luminous reflectance. Colors
of different hues have a different degree of color glow. The result of these find-
ings is that modulated luminous reflectance is not a good predictor of perceived
lightness of chromatic stimuli. Helmholtz remarked upon this effect (1860), and
it was investigated by V. A. Kohlrausch in 1935, among others, and has become
known as the Helmholtz-Kohlrausch effect (HKE). The implication is that an
effect in addition to that described by the luminous efficiency function con-
tributes to perceived lightness. The effect is often quantitatively described by
the B/L ratio, where B is the luminance or luminous reflectance of the refer-
ence stimulus, usually achromatic, and L that of the test stimulus (e.g., see
Wyszecki and Stiles, 1982). It is of considerable importance for the photome-
try of colored lights but also for the definition of a uniform color space.
Interestingly and surprisingly, in the Munsell 1915 atlas the majority of the
most highly chromatic chips have a luminous reflectance that is higher than
that of the gray with the same visually judged Munsell value. In the Renota-
tions value was defined in terms of luminous reflectance and all colors of a
given value have the same luminous reflectance value. The Re-renotations
incorporate the HKE and colors of the same value differ in luminous
reflectance. The same applies to the OSA-UCS system where the HKE also
has been incorporated. In the 1950s and 1960s important experimental studies
of the effect were made. Among these is the Wyszecki-Sanders work of 1964
with 20 observers using a visual colorimeter, and an experiment by the Com-
mittee on Uniform Color Scales of the Optical Society reported by Wyszecki
(1967). The latter experiment was to confirm or modify the findings of
Wyszecki and Sanders and involved a total of 76 observers viewing chromatic
tiles against a series of achromatic tiles. In analyzing their experimental data
in the CIE chromaticity diagram, Wyszecki and Sanders found the resulting
iso-brightness to luminance (B/L) contours to vary in irregular fashion (Fig.
5-20). The effect is weakest for yellow and strongest for purple colors. The tile
experiment of the OSA committee produced similar contours. Wyszecki and
Sanders fitted an analytical formula in the CIE chromaticity diagram that pre-
dicts the B/L ratio from the chromaticity coordinates of the color. By that
formula the correlation coefficient between experimental and calculated
results for their data was found to be 0.92. In 1991 Fairchild and Pirotta found
that the magnitude of the implicit additivity failure depends on the luminance
level of the surround.
More recently experimental results and interpretations have been reported
by Nayatani and co-workers. Their investigations revealed that the magnitude
184 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-20 Iso-brightness to luminance (B/L) contours in the CIE chromaticity diagram as deter-
mined by Wyszecki and Sanders (1964), the result of the Helmholtz-Kohlrausch effect. W iden-
tifies the white point. The effect is weakest for yellow and strongest for purple colors.

of HKE depends on the experimental situation (Nayatani et al., 1994a,b). In

1997 Nayatani offered a formula that predicted the Wyszecki-Sanders data
with a correlation coefficient of 0.89. It involves the determination of a spec-
tral function that resembles the chromatic threshold function and Evans’s
function of zero grayness. In 1998 Nayatani proposed a formula where the
spectral chromatic strength function was calculated in three sections from
Hurvich and Jameson type opponent color functions: below 500 nm it is cal-
culated as the square root of the sum of the squares of the B and R lobes of
the opponent function, between 500 and 580 nm from the G lobe, and above
580 nm from the R lobe.
In 2000 Kuehni reported analysis results for the Wyszecki-Sanders and the
Wyszecki 1967 data (Kuehni, 2000d), using a linear opponent color system. For
the former a correlation coefficient of 0.95 was obtained for predicting the B/L
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 185

ratios using a formula where the perceived brightness was calculated by

adding, by quadrant, a portion of the corresponding linear a and/or b oppo-
nent color value as follows:

YA = YC + 0.23 a if b is positive,
0.5
YA = YC + 0.20 (a 2 + b 2 ) if a is negative and b is negative,

YA = YC + 0.30 a if a is positive and b is negative. (5-11)

YC is the luminosity value of the sample. For all yellowish colors a portion of
the positive, respectively negative, lobes of the a function is added; for bluish-
greenish colors a portion of the geometrically summed two opponent func-
tions is added; for reddish-bluish colors a slightly larger portion of positive
lobes of the a function is added. To optimally predict the tile data of 1967,
larger contributions of opponent color signals in a somewhat different vector
composition was required:
0.5
YA = YC + 0.43 (a 2 + b 2 ) if b is negative,
0.5
YA = YC + 0.28 (a 2 + b 2 ) if b is positive. (5-12)

This formula predicts the committee data with a correlation coefficient of 0.88.
The corresponding effective luminosity function is compared with the CIE
luminosity function in Fig. 5-21. When inverted for comparison to the chro-

Fig. 5-21 Spectral effective luminous reflectance for the OSA-UCS Helmholtz-Kohlrausch
effect data, as modeled by the author (2000d), compared to the 10° observer ȳ function.
186 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

matic strength function, its long wave trough is found to be shifted toward
lower wavelengths. The reason for the difference of the apparent contribution
of opponent color signals to perceived brightness in the two experiments is
not clear. However, these results provide implicit support for the idea that the
opponent color system makes a contribution to perceived brightness/lightness
of chromatic colors.

Psychophysical Scaling of Chroma

In their representation in the psychological diagram, Munsell colors of equal
chroma are located on concentric circles. In the CIE chromaticity diagram,
they lie on ovoid-shaped contours with the major axis in the direction of
approximately dominant wavelength 575 nm, meaning in the yellow-blue
direction (see Fig. 7-13). There have been many efforts to find models for
chroma spacing that are in better agreement with the psychological spacing.
A significant improvement is obtained when plotting them in an opponent
color diagram. In an Adams type opponent color space, such as the CIELAB
space, the contour (see Fig. 5-24a) is improved but continues to be unbalanced
in the yellow-blue direction. In 1946 Saunderson and Milner offered an ana-
lytical solution modifying the Adams chromatic diagram by changing the
scaling of the yellow-blue axis. In an analysis in 1949 by Burnham this adjust-
ment was found to represent the best fit to the Munsell system. However, he
offered no explanation of its need.
What Munsell called chroma has been identified in more recent literature
as contrast (at constant luminous reflectance). Contrast has been scaled in
terms of contrast threshold steps (e.g, see Eskew et al., 1999). But systematic
scaling comparable to the psychological scaling discussed below is lacking.

Constant Chroma Circles

As indicated in Chapter 4, since the establishment of the Munsell Renotations
only a few systematic determinations of constant chroma around a hue circle
have been made. The Re-renotations, a modification of the Renotations for
certain findings obtained in connection with the development of the OSA
Uniform Color Scales, have chroma scales significantly different from those of
the Reenotations, and the constant chroma contour has a different shape and
angle of the major axis (see Chapter 7). In 1957–8 Nickerson, Judd, and
Nimeroff, also in preparation for OSA-UCS, developed a circle of constant
chroma samples at value 6 (Judd, 1965), as mentioned in the previous chapter.
It again differs noticeably in shape and angle of the major axis of the ovoid
from those of the Munsell system. Chroma scales are also implicit in the OSA-
UCS system, assuming that chroma can be calculated as the euclidean sum of
g and j. The formulas for g and j represent the committee’s euclidean fit to
average judgments by 49 to 76 observers of color differences between 128
nearest neighbor pairs of a triangular grid of colors. The resulting ovoid shape
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 187

is again significantly different. There is no objective argument for any one of

the four scales being most accurate. None of the four psychological chroma
scales is well represented by the CIELAB formula. The four “constant”
chroma contours are illustrated in Fig. 5-22 a and b in the balanced cone sen-
sitivity opponent diagram [5.67 (L - 1.9884M) vs. L + M - S ], indicating the
significant differences. The four scales and the scale implicit in CIELAB have
been modeled with cone sensitivity based opponent color functions. The
results are listed in Table 5-1, which also contains the coefficient of variation
in percent (COV) of the calculated chroma values. The equations indicate that
the contours differ primarily in the contribution of the b function to the a func-
tion, namely the implied degree of reappearance of red at the short wave end
of the spectrum, and weightings that differ by quadrant. UCS and CIELAB
resemble each other in their weightings of the b scale, induced by the common
cube root power modulation. The fact that the constants for CIELAB are sig-
nificantly different and the coefficient of variation is much higher indicates
that it is not a good representation for any of the four visually based contours.
As we saw in Chapter 4, indirect chroma scales derived from judgments of
the content of unique hues in Munsell chips (Indow’s work) have not resulted
in meaningful scales. As a result we have comparatively poor agreement
between various direct and indirect chroma scaling results and no good fit by
the formula currently recommended by the CIE.This applies also to the CIE94
formula whose only difference in regard to chroma is that it has an adjustment
for chromatic crispening and the basic ovoid shape of the unit difference
contour. As will be discussed later, chromatic crispening is likely limited to
comparatively small differences.

TABLE 5-1 Cone sensitivity based opponent color models fitted to “uniform” chroma
circles
Data a b
Munsell Renotations a = 0.926 [5.67(L - 1.9884M) - 0.50b] b+ = L + M - S
COV = 5.6% b- = 0.90(L + M - S)
Munsell Re-renotations a+ = 0.646 [5.67(L - 1.9884M) - 0.25b] b+ = 1.09(L + M - S)
a- = 0.807 [5.67(L - 1.9884M) - 0.25b] b- = L + M - S
COV = 1.6%
Nickerson et al. a+ = 0.953 [5.67(L - 1.9884M) - 0.20b] b =L+M-S
a- = 1.030 [5.67(L - 1.9884M) - 0.20b]
COV = 4.4%
OSA-UCS a+ = 0.908 [5.67(L - 1.9884M) - 0.40b] b+ = 1.33(L + M - S)
a- = 0.976 [5.67(L - 1.9884M) - 0.40b] b- = 0.75(L + M - S)
COV = 4.7%
CIELAB a+ = 0.676 [5.67(L - 1.9884M) - 0.25b] b+ = 1.33(L + M - S)
a- = 5.67(L - 1.9884M) - 0.25b b- = 0.65(L + M - S)
COV = 8.8%
188 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-22a Constant chroma contours determined in three different experiments in the bal-
anced cone opponent diagram: Nickerson et al. (Judd, 1965), Munsell Renotations, and Munsell
Re-renotations.

Fig. 5-22b Constant chroma contour implicit in the OSA-UCS system, in the balanced cone
opponent diagram.
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 189

TABLE 5-2 Optimal powers to linearize relations

between psychophysical and psychological chroma
scale values at or near the axes
Data a+ a- b+ b-
Munsell Renotations 0.15 0.19 0.42 0.06
COV, % 6.5 4.7 13.1 4.3
Munsell Re-renotations 0.45 0.75 0.70 0.38
COV, % 2.2 6.4 11.1 6.1
OSA-UCS 0.70 0.84 0.58 0.33
COV, % 0.3 0.4 1.8 0.2

Relation of Constant Chroma Circles at Different Chroma Levels

Psychophysical constant chroma contours may be established at different
levels of chroma. If they are established so that they are perceptually equally
spaced, the question arises as to how the psychophysical values are related for
different chroma levels at constant hue and constant luminous reflectance. As
was mentioned earlier, an important proposal was made in 1942 by E. Q.
Adams. He suggested that the CIE tristimulus values should be scaled in accor-
dance with the Munsell value function. Later the Munsell function was
replaced by a power function, and the current CIE recommendation in the
CIELAB color space and difference formula employs a cube root function.
The same power function was also used in the development of the OSA-UCS
formula. However, analysis of the Munsell Renotations, Re-renotations, and
OSA-UCS indicates that the effective optimal power is usually different from
cube root and differs by semi axis (Kuehni 2000b). Table 5-2 lists the optimal
powers to be applied to tristimulus values for the three data sets.
The results, while indicating considerable differences in the optimal powers
for the three sets of data, show a degree of relatedness of the Re-renotations
to OSA-UCS, as indeed attempted by Judd. The low COV values of OSA-UCS
are due to the systematic tiling of the j, g diagram once a formula had been
fitted to the basic experimental data (see Chapter 7 for details). Given the
uncertainty in the related uniform chroma circles and the variability in powers,
it is also here not apparent which system, if any, is accurate.

Interrelation between Chroma, Hue, and Lightness Differences

The uncertain relationship between combined hue and chroma differences, on
the one hand, and the chromatic and lightness differences, on the other, was
discussed in the previous chapter. Color space and difference formulas gen-
erally assume euclidean summation. However, the Nickerson and Indow for-
mulas (in Chapter 4) argue for linear addition. The fact that the CIELAB
formula is not a good model for any of the experimental chroma and hue scales
brings up the issue of conflated component differences calculated from that
formula. Figure 5-23 illustrates two colors A and B located on a quarter circle
190 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-23 Schematic depiction of quadrant 1 of an opponent color diagram with constant
chroma contours calculated from a formula and determined by an experiment.

in the a*, b* diagram and therefore representing a hue difference only.

However, if the true constant chroma contour is represented as indicated in
the figure, as might be possible based on the equations in Table 5-1, the dif-
ference between A and B is in reality a mixed difference consisting in part of
a hue and in part of a chroma difference. The degree of this conflation also
depends on the size of the unit hue difference along the constant chroma
contour. We cannot expect to have a reliable hue scale unless we have a reli-
able chroma scale. This fact was recognized by Judd and his co-workers in their
work in the late 1950s. A reliable chroma scale, at the same time, depends on
a reliable determination of constant perceptual lightness planes.
Additional conflation can result from the different effective powers shown
in Table 5-2. It is evident that we cannot rely on hue and chroma differences
calculated from CIELAB to have an accurate relationship to the perceptual
data of the three data sets. Conflation between lightness and chromatic dif-
ferences in CIELAB related formulas is also a fact because chromatic and
achromatic colors of equal perceived lightness are not on the same value plane
due to the Helmholtz-Kohlrausch effect. A further difficulty is that, as will be
seen Chapter 7, a euclidean global color space uniform in terms of hue dif-
ferences and in terms of chroma differences cannot be uniform in both. At the
same time it remains to be seen how uniformly and repeatably observers judge
component differences in complex small suprathreshold differences. Such
experiments are overdue.
The shape of unit difference contours in a uniform color space is by defin-
ition spherical. But that space is not automatically euclidean, as will be shown
in Chapters 7 and 8. In a euclidean chromatic diagram the unit contour has
trapezoid, or perhaps egg-shaped, rather than circular form. In the recent color
vision literature various kinds of contours have provided best fits to threshold
data in cone contrast space: superellipses, rhombi, and rectangles. For the same
 EXPRESSING PSYCHOLOGICAL SCALES IN PSYCHOPHYSICAL SPACES 191

experimental conditions often different contours provided the best fit to data
of different observers (e.g., see Sankeralli and Mullen, 1996).

Psychophysical Hue Scaling

The intent of the psychological Munsell chromatic diagram is to be uniform
in terms of hue differences. On the other hand, as seen before, in the psycho-
logical Munsell diagram unique hues do not fall on diagram axes, and these
axes are therefore not psychologically meaningful. It is useful to determine
the location of unique hues in the CIELAB diagram to learn if there is a better
agreement with the system axes than in the Munsell diagram. This is illustrated
in Fig. 5-24a and b, and we find that for the 2° observer two average unique
hues fall close to the system axes.
There are significant discrepancies for the green and red unique hues. The
CIELAB formula and some formulas recommended later for small color dif-
ferences imply equal hue angle differences for equal hue differences. CIE94
(see Chapter 6) weights only the magnitude of the calculated hue difference.
Various functions adjusting hue differences as a function of hue angle have
been introduced into other formulas, including the new CIEDE2000 (see next

Fig. 5-24a Munsell Renotation colors of value 6, chroma 8 in the a*, b* diagram of the CIE
2° standard observer and the equal energy illuminant. The letters indicate the locations of the
average unique hues determined in the experiment by the author (2001a).
192 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-24b Colors of Fig. 5-24a illustrated in the a*, b* diagram of the CIE 10° standard
observer, with locations of average unique hues.

chapter). These functions not only embody the fundamental variability (as
related to uniform hue difference) in hue angle difference but also correction
for the error added by the CIELAB formula.
We can surmise from Table 4-1 that the various described experiments have
somewhat different hue scales. We find this to be the case when comparing
hue angle differences in the optimized cone sensitivity based opponent dia-
grams of Table 5-1 around the hue circuit for the Munsell Renotations, the Re-
renotations and the Newhall data. These are the only data sets where we have
continuous consistent hue scaling around a psychologically scaled constant
chroma circle (see Fig. 5-25 for the Newhall data). The results indicate little
or no discernible agreement except for the region of hue angles 0 to 100 for
the Newhall and Re-renotation data. While the changes in hue angle differ-
ence appear to be random in the Renotations, there are systematic minima
near hue angles 75° and 275° in the Re-renotations. The Newhall data have a
systematic development in roughly the first half of the hue circuit. Overall, the
lowest level of disagreement is between the Renotations and the Newhall data.
All three sets of hue angle differences have slightly positive slopes relative to
the hue angle—Renotations 0.0099, Newhall 0.0076, Re-renotations 0.0048—
indicating a tendency for the increment to become larger as the hue angle
increases.
 COLOR MATCHING AND APPEARANCE SCALING 193

Fig. 5-25 Hue angle differences in degrees as a function of hue angle between uniform hue
steps on a constant chroma contour in the optimized cone sensitivity based opponent diagram,
Newhall data. The straight line is a linear regression line.

Partial scaling of hue circles has been done by Qiao et al. (1998) and several
color vision scientists (e.g., see Krauskopf and Gegenfurtner, 1992). It is also
implicit in small color difference data (but assuming that a total color differ-
ence can be split into components based on the euclidean model and that the
CIELAB formula describes visually uniform chroma circles). Differences in
stimulus increments needed at different hue angles may be the result of dif-
ferent specific mechanisms being responsible for hue identification and
discrimination.

5.9 COLOR MATCHING AND APPEARANCE SCALING

Color-matching functions are expressive of the response of the visual system to

the color-matching task.This task does not necessarily have a direct connection
to color appearance scaling, even though both presumably are derived from the
cone responses. The fact that CIE opponent color diagrams roughly match a
Hering type psychological diagram has been suggestive of such a connection.
As mentioned, opponent axes for the 2° observer differ from those for the
10° observer. The differences between the 2° and the 10° color-matching func-
tions are thought to be primarily due to the effect of the macula in the eye (e.g.,
see Oleari, 1999). In addition the distribution of cone types differs as a function
of field size, and the optical density of the cones may change (Pokorny et al.,
1976). Macula is a layer of matter containing yellow-colored carotenoids and
other compounds. It may have several purposes, one being protecting cell layers
194 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

of the more central area of the retina by absorbing short wavelength light. The
macula has the form of an irregular spot centered on the fovea, the most sensi-
tive region of the retina. However, the central region of the fovea is free of
macular material.There is the interesting situation of the central fovea believed
to have no S cones and being macula free.10 Informal tests, with rare exceptions,
have not resulted in any discernible hue differences when simultaneously
viewing the same colored paper with a gray mask with 2° and 10° openings.
Small hue changes have been noted in case of some complex grays and some
purples. Chroma is often slightly enhanced for small fields compared to larger
ones, perhaps as a result of increased contrast at the lower field size.11 The 2°
observer geometry results in the image being focused to a good extent on the
macula-free central fovea.The 10° observer geometry, on the other hand, results
in a considerable filtering effect by the macula expressed in the different func-
tions. Nevertheless, appearance of colored materials against a neutral surround
seems largely unaffected by field size from 2° to 10°.
From the differences between the two sets of observer functions one can
predict that metameric matches viewed in a 2° field generally will not match
in a 10° field. In regard to appearance, the “blue” system axis falls on domi-
nant wavelength approximately 477 nm for the 2° observer and approximately
470 nm for the 10° observer. It means that if unique hue for the 2° observer
falls on the negative b axis, that axis for the 10° observer represents a slightly
reddish blue. For the “yellow” axis the values are 578 nm, respectively 573 nm,
implying in case of the 10° observer a slightly greenish yellow (as demon-
strated in the OSA-UCS system). The 10° blue-yellow axis can be rotated by
a linear transformation so that it is much closer to the perceptual unique blue
and yellow. The x̄ modification required is approximately as follows:
x10 ,mod = 0.9 x10 + 0.1z10 . (5-13).
However, such a change also results in significant changes in predicted chroma
of a hue circle. The difference between the two CIE standard observers there-
fore is not just a matter of axis rotation but represents a more fundamental
change. It is evident that very specific forms of color-matching functions would
be required to reduce reflectances to geometric patterns that closely resem-
ble the geometrical pattern of the psychological space. However, it is ques-
tionable if such a transformation exists. There is evidence, to be discussed later,
that such a simple approach of combining facts of color matching with facts
of color appearance is insufficient. Thus appearance is not in good agreement
with the changes implied by the color-matching functions.

5.10 PLACEMENT OF THE RED AND GREEN UNIQUE HUES IN THE

OPPONENT COLOR DIAGRAM

The dominant wavelengths of objects seen as having unique red hue are not
spectral; their complementary wavelengths fall for most observers on the
 PLACEMENT OF THE RED AND GREEN UNIQUE HUES 195

Fig. 5-26 Spectral reflectance functions of a Munsell color chip (2.5R6/8) nearly representing
average unique red hue and a modified red. Open circles represent a simplified metamer of
2.5R6/8. Inverted triangles show the short waveband adjustment required to place the color on
the positive a* axis of the CIELAB chromatic diagram.

purple line in the CIE chromaticity diagram. This is indicated by the fact that
objects with red appearance have dominant wavelengths not only in the long
wave but also in the shortwave region. Conceptually unique red is located
where blueness and yellowness are in balance. The reflectance function of an
object resulting in unique red hue appearance in standard daylight against an
achromatic surround has a small short wave and a larger long wave loop. The
real and an idealized reflectance curve representing Munsell color 2.5R6/8,
closest in hue at that level of chroma and value to average unique red, is illus-
trated in Fig. 5-26. While for the 2° observer the tristimulus values are identi-
cal, there is a calculated color difference of 0.9 CIELAB units between them
for the 10° observer. When calculating the position of this hue in the a*, b*
opponent color diagram, we find that its hue angle is not near 0° but at 18°
(15° for the 10° observer). The Munsell hue located at 0° for both observers
is 5RP, for most observers a distinctly bluish red. Instead of near zero, the b*
value for 2.5R6/8is 12.1, indicating that there is too much implied yellowness
or not enough blueness present to balance the two. One can determine how
much higher the shortwave reflectance would have to be (illustrated in Fig. 5-
26) to result in a b* value of zero. It is relatively a 40% increase. It is possible
to shift the weight of the b+ opponent function on the spectral scale while
maintaining its total weight in such a way that b* has a considerably smaller
value. But as a result dramatic imbalances of the chromas of a Munsell hue
circle are obtained. A similar situation applies to unique green. In fact there
196 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

does not appear to be a form of color-matching functions that causes objects

of all four unique hues to fall on the system axes while maintaining near con-
stant chroma for the hue circle. This situation may point to an ultimate incom-
mensurability between color matching and color appearance. The implication
is that opponent signals of the type recorded in the LGN and visual area V4
of the brain undergo additional nonlinear changes in the “black box” between
area V4 and the appearance of colors in our consciousness.

5.11 CURVATURE OF LINES OF CONSTANT HUE BLUE COLORS

Another indication of the discrepancy between color matching and color

appearance is the curvature in a, b diagrams of lines of constant hue of bluish
colors. The curvature is evident when plotting the near-unique hue 5PB at
value 4 and several chroma levels in the a*, b* diagram. It is even more dis-
tinctly noticeable when plotting the bluish colors of OSA-UCS in the 10°
observer based a*, b* diagram. Further indication is the rotation of unit color
difference ellipses of bluish colors in counterclockwise direction in the a*, b*
diagram (Fig. 5-27). This effect can be fixed mathematically by using a hue
angle based function (see Luo, Cui and Rigg, 2001). The optimal function has
to be different for the two CIE observers.
A clockwise rotation takes place when linearly transforming the x̄ color-
matching function by subtracting a small amount of z̄ and replacing it with the
corresponding amount of the original x̄:

xmod = ( x + mx ) - mz , (5-14)
where m is a factor with a value of 0.15 or less and depends on the visual angle
occupied by the sample; that is, the optimal value for m differs for the two CIE
standard observers (Kuehni, 1999). The effect of this equation is to adjust the
relative size of the two loops of the x̄ function by slightly reducing the short-
wave and enlarging the long wave loop. We have seen above that adjustments
of this type (expressed in terms of cone sensitivity) are also important for
optimal fitting of constant chroma data. The effect is also evident in the short-
wave red loop of the OSA-UCS g function (see Fig. 7-17b). P. W. Trezona and
R. P. Parkins (1998) pointed out that the size of the two loops in the CIE color-
matching functions is somewhat arbitrary. Application of the revised function
rotates not only bluish but also (if less so) yellowish colors in the a*, b*
diagram.

5.12 MUNSELL COLORS IN THE L, M, S AND X, Y, Z SPACES AND

THE A, B DIAGRAM

For a better understanding of how the three psychophysical color spaces are
related, it is instructive to compare a selection of Munsell colors (Kuehni,
 MUNSELL COLORS IN THE L, M, S AND X, Y, Z SPACES AND THE A, B DIAGRAM 197

Fig. 5-27 Ellipses in the a*, b* diagram fitted to selected Luo and Rigg ( solid) and to the RIT-
DuPont (dashed) small color difference data. Note the rotation of ellipses near the negative b*
axis. From Melgosa et al. (1997).

2001c). For this purpose a “Celtic cross” figure is selected consisting of a hue
circle at chroma 8 and colors at increasing chroma from the neutral point of
hues falling nearest the axes of the a, b diagram at values 3, 6, and 8. It is under-
stood that the hue and chroma scaling of these colors cannot be taken as
perfect, but there is no doubt that the corresponding samples are a reasonable
approximation of a uniform hue circle and uniform chroma scales. In the psy-
chological chromatic diagram the three Celtic crosses coincide forming a
single cross. Figure 5-28 illustrates these colors in the three-dimensional psy-
chological diagram. After converting the CIE tristimulus values (adjusted to
reflect an equal energy illuminant) to L, M, S values using the Smith-Pokorny
transformation, the colors are plotted in the L, M, S space in Fig. 5-29. The
three crosses form sections through an elliptical funnel. Circles in the percep-
tual diagram are strongly elongated along the S axis in the L, S and M, S planes.
The chroma steps along the S axis become increasingly larger as S increases.
Because of the definition of Y in terms of L and M the slices of the elliptical
198 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

Fig. 5-28 Representation of Munsell colors at values 3, 6, and 8 and chroma 8, and chroma
scales on the axes (Celtic crosses) in the conceptual Munsell color space where hue, value,
and chroma scales are not matched.

funnel are not perpendicular to the equal energy axis but slanted. It is evident
that a series of constant value/constant chroma Munsell hue circles has a
complex form in the L, M, S space without a simple relationship to any of the
three dimensions.
The situation improves when plotting the same colors in the X, Y, Z space
(Fig. 5-30). Since in this space one of the dimensions is luminous reflectance,
colors of a given Munsell value fall on planes perpendicular to the Y axis. The
chroma axes of the cross are well aligned with the X and Z axes of the space.
Scaling of X against Z is required to make the contours circular. This is evident
if we plot the three crosses in a normalized linear opponent color diagram
where a = 2.272 (X - Y) and b = Y - Z (Fig. 5-31). The funnel of hue circles is
now parallel to the Y axis. In order to convert the funnel to a cylinder in agree-
ment with the psychological cylinder and at the same time create uniform
chroma scales, we must determine the applicable psychophysical function(s).
As mentioned above, in 1942 Adams proposed that the Munsell value func-
tion was applicable to all three tristimulus values, and this proposal has been
enshrined in the CIELAB formula. However, the three Munsell Celtic crosses
require power modulation that is different for the four semi axes (Table 5-2),
and additional modulation to convert the cone in tristimulus space with good
accuracy into a cylinder (more on this subject in Chapter 7)
 SUPRATHRESHOLD SMALL COLOR DIFFERENCES 199

Fig. 5-29 Actual Munsell colors approximating the Celtic crosses of Fig. 5-28 in the L, M, S
cone sensitivity space; 2° observer, equal energy illuminant. The straight line represents the
equal energy locus.

5.13 SUPRATHRESHOLD SMALL COLOR DIFFERENCES

Differences in all sets of such data, as will be seen in the next chapter, have
been judged as total differences. In nearly all cases the total differences, as cal-
culated by the CIELAB formula, involve differences in all three attributes.
The historical process of fitting psychophysical formulas to such data is
described in Chapter 6. CIELAB did not result in good correlation against
visual judgments in various data sets, and efforts since the mid-1970s usually
involved modification of CIELAB to improve correlation. The most widely
used formulas today (in the United States) are CMC and CIELAB. The CIE
has issued recommendations of new formulas in 1994 and 2001. As in the case
of hue and chroma scaling data sets, different sets of small color difference
judgments vary considerably. A formula fitted to one set of data often does
not fit another set well. Some of the reasons may have to do with
different experimental conditions, often inadequately described. Other
reasons may involve the composition of the observer pools used in the exper-
iments and, possibly, cognitive components in the judgments of the observers.
Clarification of the reasons behind the psychological variability is important
for accurate model building.
200 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

5.14 DIFFERENCE THRESHOLD MEASUREMENTS

Fig. 5-30 Colors of Fig. 5-29 in the 2° observer X, Y, Z tristimulus space, equal energy illu-
minant. The straight line represents the equal energy locus.

Fig. 5-31 Colors of Fig. 5-29 projected onto a balanced linear opponent color diagram based
on tristimulus values (2° observer, equal energy illuminant).
 DIFFERENCE THRESHOLD MEASUREMENTS 201

5.14 DIFFERENCE THRESHOLD MEASUREMENTS

As will be discussed in the next chapter early measurements of color dis-

crimination thresholds expressed in psychophysical systems were made by
Wright (1941) and by Stiles (1946). Newer measurements important for the
purposes of this book, using sophisticated computer control of visual displays
with tight controls of adaptation, are those by Krauskopf and Gegenfurtner
(1992), Strohmeyer et al. (1998), Sankeralli and Mullen (1999), and Smith,
Pokorny, and Sun (2000). In recent years threshold measurements using object
color samples have been made by Richter (1985) and by Witt (1987).
Typical measurements in vision science involve the so-called pedestal tech-
nique. A color stimulus is displayed briefly (one second or less) against a
surround that controls adaptation. A small stimulus, typically at constant lumi-
nance, is added to or subtracted from the base pedestal stimulus and randomly
displayed against it. The observer indicates if she detected a difference
between the two displayed fields. The results are now generally reported in a
Derrington, Krauskopf, and Lennie type chromatic diagram (see Section 6.17)
with its axes scaled in terms of the threshold at surround chromaticity. Among
the many results are the following:

1. Thresholds are always smallest in terms of incremental stimulus when

the stimuli being compared have luminances or chromaticities straddling
those of the surround. This is confirmation of the results of Schönfelder
(1933).
2. When measuring chromatic thresholds against a neutral surround, the
stimulus increment required for a criterion response increases with
increasing contrast between surround and the reference pedestal, except
along the system axes.

In this kind of experiment only increments along the (L + M) - S axis increase,

while those along the L - M axis do not. Close to surround the increment actu-
ally slightly declines in size compared to that exactly at surround (Sankeralli
and Mullen). Threshold contours fitted to color centers at some distance from
the surround chromaticity are ellipselike, pointing in the general direction of
the diagram origin (Krauskopf and Gegenfurtner, 1992; see Fig. 8-14). Such
findings have resulted in the idea of multiple hue detection mechanisms
throughout the chromatic plane, a crucial issue in regard to modeling of hue
discrimination. Mechanisms of such a type are believed to be of cortical nature,
but no specific neurophysiological mechanism has been proposed yet. The idea
is controversial as seen from arguments by Krauskopf (1999) and by Eskew
et al. (1999).
The outcome may depend on the outcome of another controversy.
Derrington et al. (1982) found that most parvo cells in the lateral geniculate
nucleus are tuned to their cardinal chromatic axes (not unique hues), indicat-
ing no hue angle dependent selectivity. Later, in visual area V1 of the cortex,
202 PSYCHOPHYSICAL SCALING OF COLOR ATTRIBUTES: STIMULUS AND PERCEPTION

such clear orientation of cell response is absent, even though there is a slight
tendency in this direction (Lennie, 1999). The same applies to areas V2 to V4.
De Valois et al. (1997) found that LGN cells of all types have input into later
cells with color sensitivity. This may explain the differences in threshold mea-
surement results along the cardinal axes compared to away from them. These
results point to the possibility of a rather complex hue detection mechanism
that may depend on averaging computations among many cells with specific
hue tuning.

5.15 HOW MANY COLORS CAN WE DISTINGUISH?

This is an interesting question for which several answers have been given in
the past. The first determinations of this kind addressed the limited issue of
the number of discriminable steps in the spectrum. Initial experimental
attempts to answer this question were made by E. Mandelstamm, reported in
1867. Measurements considered valid for many years were those by W.
Dobrowolsky (1872). Kries computed in 1882 from Dobrowolsky’s data 208
discriminably different steps. Based on König and Dieterici’s experiments of
1884 the number increased to 235. However, there is the issue of conflation of
hue, saturation, and brightness differences when looking at spectral colors. In
addition the results depend on the experimental setup. When correcting for
brightness differences, L. A. Jones in 1917 found 128 hue steps. Many deter-
minations of wavelength discrimination have been made since then, notably
by Wright and Pitt in 1934 and Bedford and Wyszecki in 1958, with quite
similar results. MacAdam (1947), on basis of his color-matching error data, cal-
culated a number of 250 just noticeable differences in the spectrum, not far
different from that of König and Dieterici.
But color, as we know, is not just hue but also brightness/lightness and sat-
uration/chroma. In 1896 Titchener estimated the total number of perceptibly
different colors at about 33,000. By 1939 the number, as estimated by Boring,
had risen to 300,000. Both Titchener and Boring did not distinguish between
light colors and object colors. In the same year Judd (1939) estimated the
number of perceptible object colors as 10 million.This figure has become much
quoted. In 1943, in connection with their development of a psychological color
solid (see Figs. 1-2 and 1-3), Nickerson and Newhall calculated a rounded
number of 7,500,000 at the just noticeable difference level (applicable to the
model on the right side of Fig. 1-2) and a number of 1,875,000 under less favor-
able viewing conditions (applicable to the model on the left side). This number
applies to a color solid reaching out to the object color limits.
MacAdam in 1947, based on his color matching error data of 1942, limited
himself to determining the number of perceptually different colors in a con-
stant luminance plane. Using three standard deviations of the matching error
as the JND limit, he calculated the number of colors distinguishable in a con-
stant luminance plane, determined under the conditions of his color-matching
error experiment, in a rounded number, as 17,000.
 HOW MANY COLORS CAN WE DISTINGUISH? 203

More recently the question has appeared again. In 1981 A. Hård and L.
Sivik (in contrast to the 10 million colors Judd mentioned as discriminable)
estimated the number of distinct colors that can be identified with a degree of
certainty as 10 to 20,000. In 1995 Indow estimated the number of colors that
can be discriminated from each other as approximately 7 million without pro-
viding an explanation for the estimate. In 1998, using the CIELAB formula as
a basis and assuming the limit of object color distinction to be 1 CIELAB unit
of total color difference, M. R. Pointer and G. G. Attridge calculated the
number of distinguishable object colors to be 2.28 million. They also calcu-
lated the theoretical limit of different object colors displayable on a video
monitor to be 16.78 million.
Based on a single medium gray background/surround the number must be
significantly smaller than 2.3 million because of lightness and chromatic
crispening effects. At the same time the JND limit near the surround is likely
smaller than 1 CIELAB unit. The CIE94 color difference formula predicts that
one CIELAB chroma unit at metric chroma zero is only approximately 0.2
chroma units at metric chroma 100. Unless there is a widely variable surround
the number of perceptibly different object colors is more likely about 1
million. Nevertheless, it is a remarkable number of different percepts based
on the output from three cone types.
There is, however, an additional issue.The number of 1 million or 2.3 million
applies to the situation of placing a net defining JNDs over a much finer net
defining visible stimuli. Once a starting position is defined, one can perceive 1
million other colors compared to the reference color. But the reference color
can be varied within the just noticeable difference limen around it. Applying
the JND net to a new reference point, we experience 1 million slightly differ-
ent colors. The number of colors we can experience may, after all, be more in
the range of what can be produced on a video monitor.

In this chapter it has been shown that the fitting of psychophysical models to
uncertain psychological data has added another layer of problems. In the
absence of full understanding of the human visual mechanism, such modeling
is mainly empirical. The currently neurophysiologically supported (in
macaque) opponent space at the LGN level is not in agreement with the
human psychological opponent color space. But a comparatively simple
adjustment can improve the fit. Interestingly different experimental psycho-
logical constant chroma circles can be modeled closely with only changes in
the level of this adjustment and modifications in scaling of the semi axes. It is
necessary to investigate the causes of differences in constant chroma evalua-
tions: uncontrolled surround effects, variability in the personal color-matching
functions of the observer pool, perhaps other reasons. Different sets of hue
scaling data around a constant chroma circle also have provided significantly
different results. In both cases development of reliable, replicated data seems
urgent.
 Chapter 6
Historical Development
of Color Space and
Color Difference
Formulas

A color space formula is a set of mathematical equations describing a color

space model in terms of cone sensitivities or tristimulus values and spectral
power distributions. A color difference formula is a mathematical expression
resulting in numbers proportional to perceived difference between points in
the space. Historically development of color space and color difference for-
mulas has taken a standard association between a color stimulus (as repre-
sented by a family of metamers) and a color perception as a given. Recognition
of the need for exact specification of the conditions of viewing the stimuli is
a relatively recent phenomenon. Most of the formulas discussed below betray
no such recognition. As such they need to be understood as lacking funda-
mentality and being limited to implicit or explicit assumptions/representations
of viewing conditions.

6.1 LINE ELEMENTS

The line element is the so-called first fundamental form of a regular surface.1
For small color differences it is given explicitly by the Riemannian metric

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

204
 LINE ELEMENTS 205

which determines the arc length of a curve on a surface. Connected with the
line element is the geodesic: the curve that locally minimizes the length of a
curve. It depends on the geometry of the space to which it applies: in plane
geometry the geodesic is a straight line, on a sphere a segment of a great circle
(e.g., the equator). In connection with color Wyszecki and Stiles (1982) defined
the line element as “a measure of distance in a postulated space in which per-
ceived colors are represented by points or vectors.” Historically the spaces
related to given line elements have been based on color fundamentals R, G,
and B or cone sensitivities.The differences between colors in these spaces have
been defined with increments, originally based on the Weber-Fechner law and
later on more complex relationships.

Helmholtz and Schrödinger

Interest in a stimulus intensity based model of uniform color space, a model
of outer psychophysics in Fechner’s sense, grew since the end of the nineteenth
century. In 1886 König wrote that it should be easy to construct a color mixture
diagram in a manner that equal distances in all directions would be propor-
tional to equal numbers of discrimination steps. What he had in mind is a
model that would partition the diagram in agreement with the Weber-Fechner
law. The first quantitative version was introduced by Helmholtz in 1896 based
on ideas discussed in 1891. He assumed three fundamental color vision
processes R, G, and B and the uniform applicability of the Weber-Fechner law:
dR dG dB
= = = constant, (6-1)
R G B
where dR is the increment/decrement in magnitude of the stimulus described
by the fundamental process R, and comparable for the other two stimuli.
Assuming in addition a euclidean space, we can calculate the threshold dif-
ferences as
2 2 2 0.5
ÈÊ dR ˆ Ê dG ˆ Ê dB ˆ ˘
ds = Í + + ˙ , (6-2)
ÎË R ¯ Ë G ¯ Ë B ¯ ˚
where ds is the distance between two neighboring points in the R, G, B space
representing a threshold difference. Helmholtz attempted to make his model
fit experimental data of König and Dieterici by appropriately shaping his three
visual processes with linear transformations from color-matching functions.All
of the resulting functions had two maxima, and none was in agreement with
the experimental luminous efficiency function. Helmholtz was unaware of and
did not consider contributing factors important for an accurate line element.
A more complex line element was proposed in 1920 by Schrödinger in
an attempt to correct perceived shortcomings of the Helmholtz model.
Schrödinger recognized that Helmholtz’s form of the line element results in a
luminance function that is much different from experimental functions. His
proposal has the following form:
206 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

2 2 2
2 1 È l R (dR) lG (dG) l B (dB) ˘
(ds) = Í + + ˙, (6-3)
l R R + lGG + l B B Î R G B ˚

where lR , lG, and lB are constants corresponding approximately to brightnesses

of the three fundamental processes derived from the luminous efficiency func-
tion, R, G, B are the fundamental processes as experimentally determined by
König and Dieterici. The line element obeys the Weber-Fechner law while
being additive. Schrödinger’s proposal did not receive much practical atten-
tion, perhaps because of the difficulties of executing such calculations on a
routine basis.

Stiles
In 1946 Stiles offered a version of a line element based on extensive investi-
gation of two-color thresholds. Importantly Stiles found that different Weber
fractions apply to the three visual processes rather than the common fraction
assumed by Helmholtz and Schrödinger. Stiles’s line element has the follow-
ing form:
2 2
z (R) ˘ Èz (G) ˘ Èz (B) ˘
(ds) = ÈÍ
2
dR˙ + Í dG˙ + Í , (6-4)
Î r ˚ Î g ˚ Î b ˙˚

where z(R) = 9/(9+9R) and comparably for G and B; r, g, b are proportional

to the Weber fractions, with values of 1.28 for r, 1.65 for g, and 7.25 for b, indi-
cating that the “blue” process is much less sensitive than the “red” and “green”
processes. The three processes were defined by linear transformation from the
1931 CIE color-matching functions. Stiles’s line element was tested extensively
in succeeding years and was found to provide good approximations to several
sets of visual data, including the luminous efficiency function, wavelength
discrimination, and chromatic thresholds.

Luneburg’s Line Element for Visual Space

Mention should be made here of a line element for the human visual space
developed in 1947 by the mathematician R. K. Luneburg. Visual space refers
to the space segment covered by our eyes when looking ahead. Visual space
and color space are two different aspects of our visual sense, and there is no
immediate reason for color space to be in agreement with visual space. The
fact that a similar geometry appears to apply to both situations may be coin-
cidental. Based on comparisons between physical measurements and related
judgments of distance, Luneburg determined visual space to be Riemannian
(i.e., elliptical) with constant curvature. He proposed the following line
element:
 LINE ELEMENTS 207

d a 2 + d b 2 + dg 2
ds 2 = 2
(6-5)
È 1 ˘
ÍÎ1 + 4 K (a 2 + b 2 + g 2 ) ˙˚

where a, b, g are orthogonal sensory coordinates and K denotes Gaussian total

curvature. If K = 0, the space is euclidean; if negative, the space is hyperbolic,
and if positive, it is elliptic.

Walraven, Bouman, Vos

In 1966 P. L. Walraven and M. A. Bouman offered a new proposal for a line
element of color employing photon noise methodology (De Vries-Rose
behavior; De Vries, 1948) in place of the empirical Weber-Fechner law. With
the exception of the photon noise methodology replacing the Weber-Fechner
law the proposal was identical to that of Schrödinger. Testing against new
wavelength discrimination and other data did not provide fully satisfactory
results.
Walraven continued work with a new co-worker, H. Vos, and they devel-
oped the Vos-Walraven line element (Vos and Walraven, 1972, 1991). The
model underwent several modifications. In its original form it was based on a
model of retinal color processing which included a Helmholtz type cone
absorption step taking input of the R, G, B cones at a ratio of 32 : 16 : 1, with
cone response compression. In a summation zone the compressed R and G
signals form the yellow signal, and all three together the luminance signal.
Separately, R and G signals are balanced to form an antagonistic Hering type
red-green signal. Similarly the yellow signal is balanced against the B signal,
forming the yellow-blue output signal. Finally the luminance signal is balanced
against a surround luminance signal, resulting in a brightness contrast signal.
Vos and Walraven concluded that opponent processing does not play a role at
the threshold level and their line element is essentially of the Helmholtz type.
It is defined as follows (Vos, 1979):
2 2
Ï ¸ Ï ¸
Ô Ô Ô ÔÔ
2 Ô dR Ô Ô dG
(ds) = Ì 0.5 ˝
+ Ì 0.5 ˝
ÔÈ Ê R R2 ˆ ˘ Ô ÔÈ Ê G G2 ˆ ˘ Ô
ÔÓ ÍÎRË 1 + R0 + R 2 ¯ ˙˚ Ô˛ ÔÓ ÍÎGË 1 + G0 + G 2 ¯ ˙˚ Ô˛
1 1
2
Ê ˆ
Á dB ˜
+Á 0 .5
˜
ÁÈ Ê B B2 ˆ ˘ ˜
Á ÍB 1 + + ˜
ËÎ Ë B0 B12 ¯ ˙˚ ¯ (6-6)

where R, G, B are the cone signals at the ratio of 32 : 16 : 1, subscript 0 indi-

cates the number of quanta for which saturation occurs, and subscript 1 the
208 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

number of quanta for which supersaturation occurs. The model thus can
account for various levels of saturation of the cone system and in this manner
can predict, in addition to color discrimination at the threshold level, the
Bezold-Brücke and the Stiles-Crawford effects. While the model is a modified
version of the Helmholtz line element, it considers the ratio of cones that gives
rise to different Weber fractions for the three cone types as well as saturation
and supersaturation effects, making it not just applicable to a middle range of
luminance but to the complete range.
The Helmholtz, Schrödinger, Stiles, and Vos-Walraven line elements were
believed to be more or less uniform in terms of thresholds. However, as will
be shown in Chapter 8, they are regular color spaces. In his paper Schrödinger
suggested that geodesics, the lines indicating the paths of smallest numbers of
threshold differences, are lines of constant hue. MacAdam originally deter-
mined geodesics based on his color-matching error data mechanically, by
stretching threads across a model representing his ellipses (MacAdam, 1981).
Geodesics can be calculated, for example, by using linear programming (Jain,
1972). For an example of constant hue and saturation geodesic lines in the CIE
chromaticity diagram calculated from MacAdam’s 1965 geodesic chromatic-
ity diagram (MacAdam, 1981), see Fig. 6-1. With known hue, chroma, and light-
ness geodesics, it is possible to develop formulas transforming nonuniform
cone sensitivity or tristimulus spaces into a space uniform in terms of the
assumptions of the model involved.

6.2 PROJECTIVE TRANSFORMATIONS

Another approach to a uniform color space was based on the idea that the
CIE 1931 x, y chromaticity diagram could be linearly transformed to result in
a modified diagram in which distances were proportional to visual distances.
In 1932 D. B. Judd offered an early version representing his own threshold and
other published data (Fig. 6-2) and based on color-matching functions recom-
mended by the Optical Society of America in 1922. The diagram contains
radial lines of constant dominant wavelength and ovoids of constant colori-
metric purity. In 1935 Judd published a modified version of a uniform chro-
maticity diagram in the form of a Maxwell triangle (Fig. 6-3). In a different
reference frame the diagram became in 1960 the basis for the CIE u, v
diagram. In 1935 Judd also introduced the symbol DE to denote a color dif-
ference. In the following year Judd published a graph of the CIE 1931 chro-
maticity diagram with ellipses that represent circles of equal size in his 1935
diagram (Fig. 6-4). These ellipses are intended to represent uniform threshold
color differences, enlarged 100 times. They illustrated for the first time explic-
itly the perceptual nonuniformity of the CIE chromaticity diagram.
In 1937 MacAdam modified Judd’s 1935 diagram with simplified coeffi-
cients resulting in a rectangular coordinate diagram. In 1939 F. C.
Breckenridge and W. R. Schaub developed the rectangular uniform chro-
maticity scale diagram (RUCS), a transformed version of Judd’s 1935 diagram.
 PROJECTIVE TRANSFORMATIONS 209

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 6-1 Geodesic lines of constant hue and chroma in the CIE chromaticity diagram, calcu-
lated from MacAdam’s 1965 geodesic chromaticity diagram (Fig. 6-9). Solid lines: constant hue;
dashed lines: constant saturation. From MacAdam (1981).

In the same year Judd, Scofield, and Hunter proposed the a, b diagram and
a color space related to it. In 1942 this linear transformation of the CIE
chromaticity diagram became the basis of the National Bureau of Standards
(NBS) formula, with the difference units designated as NBS units or judds
(after D. B. Judd).2 The formula is as follows:
0.5

{[
DE = fg 221Y 0.25 (Da 2 + Db 2 )
0.5 2
] + [kD(Y 0.5 )]
2
} , (6-7)

where fg is a factor adjusting for glossiness and is defined as fg = Y/(Y + K),

with K usually taken as 2.5; k, having normally a value of 10, adjusts the light-
ness difference to the chromatic difference

2.4266 x - 1.3631y - 0.3214

a= ,
1.0000 x + 2.2633 y + 1.1054

0.5710 x + 1.2447 y - 0.5708

b= . (6-8)
1.0000 x + 2.2633 y + 1.1054
210 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-2 Judd’s projective transformation of the CIE chromaticity diagram of 1932 meant to
result in proportionality of distances with visual distances. The numbers on the spectral trace
represent wavelength in nanometers.
 PROJECTIVE TRANSFORMATIONS 211

Fig. 6-3 Judd’s 1935 uniform chromaticity diagram embedded in the Maxwell triangle. On the
dashed curve are colors of a black body at the indicated absolute temperatures in Kelvin.
Compare with Fig. 2-29.

In 1944 D. Farnsworth proposed a simplified transformation. It provides for

minimum deviation from circularity of Munsell hues at chroma 10 and value
5, centered on illuminant C. In 1959 Y. Sugiyama and T. Fukuda optimized a
transformation such that both the Munsell system and the MacAdam ellipses
were reasonably well represented. In 1963 Wyszecki proposed a transforma-
tion formula that became the basis of the CIE 1964 (U*V*W*) system:

U * = 13 W * (u - u0 ),

V * = 13 W * (v - v0 ),

W * = 25 Y 1 3 - 17, (6-9)
212 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-4 CIE chromaticity diagram with ellipses that represent circles of equal size in Judd’s
1935 diagram. From Judd (1936).

where u = 4 X/ (X + 15Y + 3Z), v = 6 Y/(X + 15Y + 3Z), X, Y, Z are CIE tris-

timulus values, u0, v0 are the values of the variables u and v for the achromatic
color at the origin of the U*, V* diagram. The total color difference was cal-
culated as the square root of the sum of the squares of the differences in the
three dimensions. This formula has been superseded by the CIELUV formula.
 FITTING MODELS TO THE MUNSELL SYSTEM 213

Fig. 6-5 Munsell Renotation colors at value 5 as represented in the CIELUV projective trans-
formation formula. From Mahy et al. (1994).

The development of projective transformation came to a conclusion with

the recommendation in 1976 by the CIE of the 1976 L*u*v* transformation
(CIELUV). This is a minor modification of MacAdam’s 1937 diagram (expan-
sion of v by a factor 1.5). A cross section of the Munsell space at value 6 as
represented by this formula is shown in Fig. 6-5.
Projective transformation employs the following basic formulas:

Ê a11 x + a12 y + a13 ˆ Ê a21 x + a22 y + a23 ˆ

x' = , y' = , (6-10)
Ë a31 x + a32 y + a33 ¯ Ë a31 x + a32 y + a33 ¯

where x¢, y¢ are the transformed chromaticity coordinates and x, y are the CIE
chromaticity coordinates; axx are transformation coefficients. Several of the
formulas are in a simplified form requiring only five coefficients. The coeffi-
cients of the above and other linear transformation formulas are given in Table
I (6.4) of Wyszecki and Stiles (1982). Linear transformations have the advan-
tage, and are thus of interest to optical and lighting engineers, that the system
remains additive and mixtures of two lights fall on a straight line connecting
their locations in the diagram. The CIELUV formula will be further discussed
below.

6.3 FITTING MODELS TO THE MUNSELL SYSTEM

Another avenue pursued involved attempts to develop a uniform color space

model from analysis of the structure of the Munsell system. This system had
been extensively investigated by the Subcommittee on Uniform Spacing of
Munsell Colors of the Optical Society of America. It published a final report
in 1943, containing colorimetric specifications of the smoothed Munsell Reno-
214 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

tation colors. In this form the Munsell system was taken to represent a good
approximation of a visually uniform psychological color space.
In 1936 Nickerson proposed the first formula with the goal of predicting
perceptually equal color differences, the Nickerson Index of Fading (see equa-
tion 4-1). The formula has a purely psychological basis, since no physical mea-
surements are involved in its application. Its immediate purpose was to
provide a measure of fading by light of dyed textiles. Component differences
were combined by simple addition. The formula recognized the dependence
in a polar system of the hue difference on chroma and also indicated that in
the Munsell system unit differences in the three attributes are of different per-
ceptual magnitude. As a result such a system is not uniform, as it doesn’t rep-
resent distances in all directions proportionally to differences in its euclidean
space (see Chapter 7).
In 1942 Adams plotted Munsell colors at several values in what he called a
“chromance” diagram, representing a linear opponent color diagram norma-
lized to the equal energy point. The two dimensions were defined as X - Y and
Z - Y. By applying the Munsell-Sloan-Godlove lightness formula (a power 0.5
modulation; see Chapter 5) not only to the luminous reflectance value Y but
also to the other two tristimulus values, Adams converted the chromance
diagram to a chromatic value diagram. Adams interpreted the subtractions as
representations of his 1923 theory of color vision that took the color-matching
functions to be cone responses. In this theory Adams had assumed an inhibitory
effect of the output of the postulated Y cone on the outputs of the X, respec-
tively Z cones. He represented this inhibitory effect with subtractions. He called
the outputs VX, VY, and VZ.Adams calculated chromatic values of a high chroma
Munsell hue circle and found those related to Z to be larger than those related
to X and Y.To have scales of maximum value 10 in all three dimensions he pro-
posed to multiply the VX values by 1.90 and the Vz values by 0.72. The result of
these operations showed reasonably circular contours in the chromatic value
diagram for Munsell colors of the 1929 Book of Color and the preliminary
smoothed values of the committee (see Fig. 5-12). Adams did not propose a
color difference formula based on his VX ,VY, VZ space, but assuming that it is
a euclidean space, such a formula is obvious. Adams’s chromatic value system
proved to be influential in coming years.
Aside from the Munsell data, in flux until the release of the Renotations
in 1943, there were few object color sample sets with statistically supported
visual data. In 1941 Balinkin reported on a set of five pale green tiles. One pair
of tiles was designated as the standard difference, and sixty observers estimated
the differences of all possible pairs against the reference pair. The data were
used in several studies of color difference formulas. In 1944 Nickerson and her
co-worker Stultz investigated the usefulness of color difference calculation for
color quality control work. The visual data consisted of category judgments
(five categories) by twelve observers of painted textile samples, primarily in
two color regions: yellowish brown and olive green. In this study they used,
among others, a formula based on Adams’s chromatic value space as follows:
 JUDD’S MODEL OF MÜLLER’S THEORY OF COLOR VISION 215

2 0.5
[ 2 2
DE A = (DVY ) + {D(VX - VY )} + {0.4 D(VZ - VY )} ] , (6-11)

where VX ,VY, VZ are the Adams chromatic values of the samples, now inter-
preted in terms of the revised Munsell value function. This formula became
known as the Adams-Nickerson color difference formula and proved of endur-
ing value. In their evaluations Nickerson and Stultz found the formula only
marginally better than formulas derived by Judd but considerably easier to
calculate. The experiments also indicated considerable individual variability in
judgment.
In 1946 J. L. Saunderson and B. L. Milner proposed a modification of the
Adams chromatic value system to obtain closer agreement with the Munsell
system. Contours of constant chroma are somewhat eccentric in the Adams
chromatic diagram, and the Saunderson-Milner solution corrected for the
eccentricity using an empirical trigonometric method. The Saunderson-Milner
color space model was described as the “Zeta” space, based on their use of the
Greek letter. It is defined as

z 1 = (VX - VY )(9.37 + 0.79 cos Q),

z 2 = kVY ,

z 3 = (VZ - VY )(3.33 + 0.87 sin Q), (6-12)

where Q is the angle calculated from tanQ = 0.4(VZ - VY)/(VZ - VY), and k is
a constant depending on the observation conditions. Color differences are
calculated as the square root of the sum of the squares of the differences in
the three z values. A somewhat different procedure with a comparable effect
was proposed in 1952 by Godlove.
Assuming that two concentric circles of five equally spaced Munsell hues
are a good representation of psychologically uniform space, Burnham in 1949
investigated the performance of ten formulas (including the CIE x, y and x, z
diagrams) and found the Saunderson-Milner Zeta space to perform best.
However, all of the formulas resulted in deviations that were in visual terms
statistically significant.

6.4 JUDD’S MODEL OF MÜLLER’S THEORY OF COLOR VISION

In 1949 Judd began to develop a model of G. E. Müller’s theory of color vision

(to be discussed in more detail later in this chapter). Development continued
into the 1960s and the model was influential in Friele’s treatment of
MacAdam’s data resulting in the FMC I and FCM II formulas to be discussed
below. The Müller-Judd space is further discussed in Section 6.17.
216 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

6.5 COLOR DIFFERENCE THRESHOLDS AND MATCHING ERROR

Threshold differences were first investigated along the spectrum because exact
setting of spectral differences was technically relatively easy. However, spec-
tral differences are complex in visual terms since every difference is composed
of hue, saturation, and brightness components. Wavelength discrimination
began to be investigated at the turn of the twentieth century. A classical inves-
tigation is that by W. D. Wright and F. H. G. Pitt (1934). Wright subsequently
also investigated threshold differences along straight lines in the CIE chro-
maticity diagram. The results of both investigations are illustrated in Fig. 6-6
(Wright, 1941). Wright’s important data were overshadowed by MacAdam’s
extensive color matching error data of a single observer, published a year later
(1942). As discussed in Chapter 3, in MacAdam’s work two hemi-fields were
displayed against a dark surround in a specially constructed colorimeter. In
one-half a standard color was displayed. The color of the second half could be
adjusted by the observer along straight lines passing through the standard
color in the CIE chromaticity diagram, in a constant luminance plane. Using
a single knob, the observer adjusted the color of the test field by the method
of adjustment until a visual match between the two hemi-fields was achieved.
The match was approached along a given line from both sides. Color matches
along several lines were repeatedly set for each of 25 standard colors. From
the visual data (some 20,000 observations) MacAdam calculated the standard
error of color matching for his single observer. The result, fitted with ellipses,
is illustrated in Fig. 6-7. MacAdam also determined that the threshold differ-
ences around his standard colors were approximately twice the standard
deviation of the color-matching errors. Subsequently comparable but three-
dimensional contours involving also brightness differences were determined
for additional observers, and the results indicated that observers vary signifi-
cantly in this task (Brown and MacAdam, 1949; Brown, 1957). The MacAdam
ellipses rapidly became a key set of data used as test data for line elements
and color difference formulas. Color-matching error was explained in 1949 in
terms of cone activity by Y. LeGrand (see Chapter 8). The ellipses obtained
good confirmation in an experiment by R. M. Boynton and N. Kambe (1980).
Additional determinations of color-matching error were made by G. Wyszecki
and G. Fielder (1971). Determinations of achromatic and chromatic thresholds
using industrially relevant conditions were performed by K. Richter (1985)
and by Witt (1987, 1990). For a comparison of such data, see Chapter 8. As
mentioned in the previous chapter, thresholds and color-matching error data
of limited groups of colors and using various methodologies have been per-
formed in recent years by several researchers.

MacAdam’s Empirical Line Element

Using a method suitable for describing line elements independent of the form
of the implied space (i.e., applicable to euclidean and non-euclidean space),
 COLOR DIFFERENCE THRESHOLDS AND MATCHING ERROR 217

Fig. 6-6 Perceptually equal steps between spectral colors as well as colors in the interior of
the chromaticity diagram, as determined by Wright (1969).

MacAdam calculated gik values from the following equation that describes his
ellipses in geometrical terms:
2 2 2
(ds) = g11 (dx) + 2 g12 dxdy + g 22 (dy) = 1, (6-13)
218 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 6-7 Color matching error ellipses of observer PGN for 25 colors at the center of the
ellipses, enlarged 10 times, in the CIE 2° chromaticity diagram. From MacAdam (1942).

where ds is the distance between the center of an ellipse and a point on its
contour and x, y are CIE chromaticity coordinates. MacAdam drew interpo-
lating lines connecting his data points. Results are shown in Fig. 6-8a–c.
Knowing the gik values for a particular location in the chromaticity diagram
by reading them from tables or graphs with interpolation between neighbor-
ing values makes possible the calculation of color differences using eq. 6-13
and applying the square root. Alternately, the chromaticity diagram could be
modified locally to convert the ellipses to circles of equal size. Sets of charts
that simplified the calculation of color differences by this method have been
available in the 1950s and 1960s. From the results of such calculation it was
apparent that linear transformation of the CIE chromaticity diagram does not
yield a uniform chromaticity diagram in which the MacAdam ellipses form
circles of equal size.

By 1950 the situation presented itself as follows: On the one hand, there were
line elements, either derived from theoretical considerations using best esti-
 COLOR DIFFERENCE THRESHOLDS AND MATCHING ERROR 219

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

(a)
Fig. 6-8a–c Contours of the g11, 2g12, and g22 ellipse parameter functions in the CIE chro-
maticity diagram for the ellipses of Fig. 6-7.

mates of cone sensitivity functions, or empirically as in the case of the

MacAdam ellipses. On the other hand, there were linear transforms of the CIE
chromaticity diagram as well as the Adams chromatic value diagram and mod-
ifications thereof in attempts to predict the Munsell data. In terms of psy-
chophysics there were thus color space models based on accumulation of
threshold differences (local psychophysics extended to global psychophysics)
and models based on global scaling data.

Color Difference Formulas Derived from the MacAdam Data

In 1961 L. F. C. Friele began devising a formula with which MacAdam’s ellipses
could be described with good accuracy (Friele, 1961, 1965). The formula was
slighly modified and simplified by MacAdam and later by K. D. Chickering
and became known as the Friele-MacAdam–Chickering or FMC I formula.
220 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

(b)
Fig. 6-8a–c (Continued)

The Roman numeral distinguished it from a later version, FMC II, with a dif-
ferent treatment of lightness. (For an informative discussion of the develop-
ment of these formulas see MacAdam 1981.) Friele based his approach on the
three-stage color vision theory by G. E. Müller (1930). The CIE tristimulus
values were converted to cone sensitivity functions P, Q, and S. Opponent
color differences and lightness differences were calculated from these in two
steps.
FMC II formula
P = 0.724 X + 0.382Y - 0.098Z ,

Q = -0.48 X + 1.37Y + 0.1276Z ,

S = 0.686Z ,
(QDP - PDQ)
DC rg = 0.5
,
(P 2 + Q 2 )
SDL1
DC yb = 0.5
- DS,
(P + Q 2 )
2
 COLOR DIFFERENCE THRESHOLDS AND MATCHING ERROR 221

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 6-8a–c (Continued)

(PDP + QDQ)
DL1 = 0.5
,
(P 2 + Q 2 )
0.279 DL1
DL2 = ,
a
2 2 0.5
ÈÊ DC rg ˆ Ê DC yb ˆ ˘
DC1 = Í + ˙ ,
ÎË a ¯ Ë b ¯ ˚
0.0000173(P 2 + Q 2 )
a2 = ,
[1 + 2.73P 2Q 2 (P 4Q 4 )]
b 2 = 0.0003098(S 2 + 0.2015Y 2 ),

DC = K1 DC1 ,

DL = K 2 DL2 ,
222 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

K1 = 0.054 + 0.46Y 1 3 ,

K 2 = 0.465K1 - 0.062. (6-14)

The total color difference is calculated from the chromatic and the lightness
difference as

0.5
[ 2
DE = (DC ) + (DL2 ) ] . (6-15)

The two K functions have the purpose of adjusting the size of the implied
ellipse as a function of luminous reflectance and to adjust the lightness value
to be in reasonable agreement with Munsell lightness. They are shown above
in the simplified form proposed by MacAdam.
A uniform (in terms of color-matching error) color space model is
implicit in this difference formula, but it has not been stated explicitly.
The formula is an elaborate fitting of the MacAdam ellipses in a cone-based
opponent color framework. It represents a compromise in regard to size of
the ellipse as a function of luminous reflectance. MacAdam had found that
the color-matching error ellipses were affected only to a small extent (less
than 20%) by changes within reasonable levels in luminosity. The K1 function
made them change in size in line with cube root compression. In his work
with Brown, MacAdam had determined that ellipsoids generated from
threshold determinations including luminance differences had one of
their three axes parallel to the luminance axis. In the Adams chromatic
value space, on the other hand, corresponding ellipsoids are tilted toward
the neutral point of the chromatic diagram. One of the advantages of the
FMC formula is that it implicitly adjusts for the Helmholtz-Kohlrausch
effect.

Nonlinear Transformation of the CIE Chromaticity Diagram

MacAdam continued efforts to represent his ellipses in a chromaticity diagram
in which they would appear as circles of equal size. He abandoned a cone sen-
sitivity based solution and concentrated on mathematical methods. Using
at first a paper and scissor method and then stepwise linear regression he
developed a geodesic chromaticity diagram, illustrated in Fig. 6-9.
Using four linear transforms of the CIE chromaticity diagram, MacAdam,
1965 proposed the following chromaticity coordinates for this diagram:

x = 3751a12 - 10 a14 - 520 b12 + 13295b13 + 32327a1 b1 - 2549212 b1

-41672 a1 b12 + 10 a13 b1 - 5227a10.5 + 2952 a10.25

h = 404 b2 - 185b22 + 52 b23 + 69a 2 (1 - b22 ) - 3a 22 b2 + 30 a 2 b23 , (6-16)

 COLOR DIFFERENCE THRESHOLDS AND MATCHING ERROR 223

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 6-9 MacAdam’s geodesic uniform chromaticity diagram, 1965. Lines of constant value of
chromaticity coordinates x and y are drawn in the interior.

where

a1 = 10 x (2.4 x + 34 y + 1) ,

a2 = 10 x (4.2 y - x + 1) ,

b1 = 10 y (2.4 x + 34 y + 1) ,

b2 = 10 y (4.2 y - x + 1) .

In evaluations (to be discussed below) against a growing number of sets of

visual color difference data involving object color samples, formulas based on
MacAdam’s ellipses performed relatively poorly, and the development of
color space and color difference model formulas generally proceeded along
different routes.
More recently a nonlinear transformation was proposed by C. Oleari
(2001). It uses an angular transformation of the CIE chromaticity diagram,
with dilatation, to form the d, u chromaticity diagram in which the
MacAdam ellipses approach circles of equal size. When this diagram is applied
to the CIE 10° observer and the OSA-UCS colors are plotted in it, they
are aligned along slanted lines which Oleari believes are due to the dif-
ference in the two and ten degree observers, which he corrected with an addi-
tional dilatation.
224 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

6.6 FURTHER DEVELOPMENT OF FORMULAS BASED ON

OPPONENT COLOR SYSTEMS

Realizing the opportunities arising from instrumental quality control of

colored materials, R. S. Hunter began to build filter tristimulus colorimeters
in 1948. These were relatively inexpensive instruments that could determine
CIE tristimulus values directly and with good accuracy from four measure-
ments per sample. Hunter, in addition, provided for output from the instru-
ments directly in terms of color difference values. Hunter proposed the
following formulas for this purpose:

L = 10Y 0.5 ,

17.5(1.02 X - Y )
aL = ,
Y 0.5
7.0(Y - 0.847Z )
bL = , (6-17)
Y 0.5

where X, Y, Z are the CIE tristimulus values for illuminant C taken as per-
centages. The total color difference was given as the square root of the sum of
the squares of the differences in L, aL, and bL. These instruments were quite
popular and continued to be used into the 1980s. Kuehni showed that unit
chromatic contours derived from the formula were not in good agreement with
ellipses fitted to small color difference data (1982).
The calculation of the Munsell value function from luminous reflectance
with the quintic formula was troublesome, and efforts were made to find a
simpler solution. It was found that properly scaled cube roots of luminous
reflectance resulted in good agreement with the quintic formula. In 1958 L. G.
Glasser, A. H. McKinney, C. D. Reilley, and P. D. Schnelle proposed a cube
root model of color space:

L* = 25.29G1 3 - 18.38,

a* = Ka (R1 3 - G1 3 ),

b* = Kb (G1 3 - B1.3 ), (6-18)

where R = 1.02X, G = Y, B = 0.847Z, Ka = 106.0, and Kb = 42.34. For closer

agreement with the Munsell data the Ka and Kb constants were adjusted by
quadrant. The total color difference was calculated as the square root of the
sum of the squares of the differences in L*, a*, and b*. With square and cube
root tables and/or then modern electronic calculators, color differences from
these formulas could be calculated comparatively rapidly.
In the 1960s, then, there were several competing formulas that were used
 NEW SMALL COLOR DIFFERENCE DATA 225

in industry to assess quality of coloration. Among these were the FMC II, the
Adams-Nickerson or the cube root, Saunderson-Milner, CIE 1964, and the
Hunter L,a,b formulas.

6.7 NEW SMALL COLOR DIFFERENCE DATA

The unsatisfactory situation in regard to industrial color quality control using

formulas available by 1960 resulted in several new sets of visual data and con-
siderable activity in testing these data and developing formula modifications.
The following sets of data were developed from the 1950s until 1980:
1. Davidson and Friede (1953). Data consisted of 287 pairs of wool textile
samples around 19 standards, evaluated by 8 observers in terms of
acceptability as a production match to the standards. The results were
expressed as acceptability percentages.
2. Robinson (1969). Data consisted of a single blue paint standard with 31
samples around it evaluated by 132 observers with the results expressed
as acceptability percentages.
3. Thurner and Walther (1969). The data consisted of 500 textile samples
around 27 standards, assessed for acceptability by 16 textile colorists.
4. Kuehni (1971a). 113 samples around three standards, pigments on
polyester/cotton fabric, were assessed for acceptability by 10 observers.
5. Kuehni/Metropolitan Section (1971b). Data consisted of 180 textile
samples around 10 standards, assessed for acceptability by 16 observers.
6. HATRA (Jaeckel 1975). 854 textile samples around 12 standards were
assessed for acceptability by 24 to 32 observers.
7. MMB (Morley, Munn and Billmeyer 1975). 555 pairs of painted samples
around 19 standards were assessed by 20 observers according to four cat-
egories from no difference to very large difference. Psychometric scales
were constructed from the visual data.
8. VVVR (Friele 1978). 20 glossy painted samples each around 10 stan-
dards were assessed by 14 observers using a pair-comparison method and
by 15 to 25 observers using the acceptability method. Psychometric scales
were constructed.
9. ISCC (Kuehni and Marcus 1979). 180 samples around six standards were
assessed, four consisting of matte paint on cardboard and two of dyed
textiles. They were assesses by 26 to 37 observers using two assessment
methods: (a) subjective estimates on a scale with a maximum value of
10; (b) acceptability as a commercial match. Psychometric scales were
constructed from the subjective estimates.
Most of these data sets were established by industrial specialists interested in
a color difference formula for quality control purposes with a reliability at
least equal to that of an average observer.
226 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

In 1970 the Society of Dyers and Colourists in England recommended the

use of the Adams-Nickerson formula (called ANLAB) for application in the
textile industry based on extensive evaluations of formulas against various sets
of visual data. It had been shown to be equal or marginally better than a dozen
other formulas under consideration. (In 1973 Kuehni found that 13 different
formulas were in use in U.S. color-related industries.) Levels of correlation
between visual and calculated data for all of these, including ANLAB, were
unsatisfactory. Formulas performed poorer than the average observer. D.
Strocka showed in 1971 that a simple formula based on circles of equal size
in the CIE chromaticity diagram and a simple lightness formula resulted in
levels of correlation comparable to that obtained with the best formulas.
Looking for causes of the low level of correlation between visual data and
formula, K. MacLaren (1971) used multiple linear regression in the directions
of lightness, metric chroma, and hue angle. He was able to show that the
magnitude of perceptually equal color differences, when calculated with
formulas such as Adams-Nickerson, increased with increasing metric chroma
and lightness.

6.8 ELLIPSE AND ELLIPSOID FITTING

Strocka’s results with the circular formula were puzzling but indicated that the
actual unit difference contours in the CIE chromaticity diagram could not be
circles, since the correlation with the circular formula would have been greater.
The implication was that the ellipsoidal contours implicit in a formula such as
Adams-Nickerson did not match the ellipsoidal contours implicit in the visual
data well, perhaps in direction as well as in size. In 1971 Kuehni graphically
fitted unit ellipses in the CIE chromaticity diagram to various sets of visual
data (Fig. 6-10; Kuehni 1971a). The results indicated that the major axes of
the resulting ellipses were usually tilted 20° to 30° clockwise compared to
the MacAdam ellipses and that ellipses tended to increase in size as chroma
increased (as McLaren had found).
Kuehni asked MacAdam to fit the parameters of his xi-eta equation (see
above) to fitted ellipses. The resulting, then unpublished, equations are as follows:

x = -4067a 2 - 133a 4 - 1675b 2 - 38280 b3 + 46479ab - 12064 a 2 b

- 2806ab 2 - 37a 3 b + 843a 0.5 - 10254 a 0.25 ,
where a = 10 x (2.5x + 34 y + 1) and b = 10 y (2.5x + 34 y + 1) ;

h = 343b - 125b 2 - 41b3 + 31a + 41a 3 + 4 a 4 - 4 ab 2 - 75a 2 b 2

- 85a 2 b + 137ab3 , (6-19)

where a = 10 x (4.3 y - x + 1) and b = 10 y (4.3 y - x + 1) .

 CONTROVERSIES OF DETAIL 227

Fig. 6-10 Ellipses graphically fitted to small suprathreshold color difference data by Robinson
1962 (Rob), and by Davidson and Friede 1953 (A–T). From Kuehni (1971a).

The lightness scale used together with the formulas above was the cube root
scale. The combined formulas resulted in a correlation with visual data sig-
nificantly higher than the cube root and ANLAB or the circle formula.
In 1972 Kuehni refined the ellipse fitting method by systematically vary-
ing ellipse parameters until highest correlation for the samples around a
standard were obtained. In 1975 R. M. Rich, F. W. Billmeyer, and W. G. Howe
employed a computer algorithm using the maximum likelihood function to fit
ellipses to visual data with greater statistical validity. Mathematical ellipse or
ellipsoid fitting has since become a standard tool of investigation of color
differences.

6.9 CONTROVERSIES OF DETAIL

A number of controversies abounded in the late 1960s to early 1970s.

One of these involved the question of change in ellipse size as a function
of lightness. MacAdam’s findings using a visual spectrophotometer had, as
mentioned earlier, indicated only small changes in the ellipse’s size with
changes in luminance. Formulas, such as Adams-Nickerson, on the other
hand, implied significant changes in unit contours as a function of lightness.
228 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

As shown in Section 8.7 the final answer to this question has not been found
yet.
Similarly Brown and MacAdam had found that the third axis of color-
matching error ellipsoids was aligned parallel with the luminance axis. For-
mulas such as Adams-Nickerson have implied third axes that are tilted toward
the white point of the color space. Work based on small suprathreshold dif-
ferences reported by W. Schultze and L. Gall in 1971 indicated no tilt of the
ellipsoids. The comparative success of power modulation-based formulas indi-
cated that tilt is appropriate.
The dependence of the ellipse’s size in a constant luminous reflectance
plane on metric chroma has already been mentioned. In addition there was
the controversy of perceptibility judgments versus acceptability judgments.
While in perceptibility experiments presumably purely psychophysical judg-
ments are obtained (but see Chapter 3), acceptability experiments can, and
sometimes do, include additional cognitive overlays. Perceptibility experi-
ments are usually difference magnitude estimations against a reference pair.
Acceptability experiments also involve difference magnitude estimation,
however, against an internal standard of acceptability in a commercial situa-
tion. Biases in the latter case are possible based on specific situations; that is,
in a chromatic diagram the limit contour of acceptable color differences may
not be positioned symmetrically around the standard. E. Allen and B. Yuhas
(1984) and, more recently, Berns (1996) have shown how this situation can be
mathematically treated. Another issue is that acceptability tolerances for iden-
tically colored materials may vary significantly depending on the context.
However, experiments have shown that tolerance contours from acceptability
judgments and unit difference contours from perceptibility judgments of the
same sample pairs, when determined in the absence of a specific context, are
symmetrical; that is to say, acceptability judgment in that situation is guided
by perceptibility (Kuehni, 1975; McLaren, 1976; Mahy, Van Eycken, and
Oosterlinck, 1994).

6.10 DEPENDENCE OF CALCULATED COLOR DIFFERENCE

ON METRIC CHROMA

Multiple linear regression as well as fitting of ellipses and ellipsoids had clearly
indicated a dependence of the chromatic differences (in the equal luminance
plane) on metric chroma. In 1972 Kuehni proposed a modification to the
Glasser et al. (see above) cube root color difference formula that adjusted the
size of the calculated total color difference as a function of the radial differ-
ence of the standard from the illuminant point in the CIE chromaticity
diagram:

2 0.5
[ 2
DE = (DC ) + (DL) ] , (6-20)
 THE CIE 1976 L*a*b* AND L*u*v* SPACES 229

where
2 0.5
[ 2
DC = (Da) + (Db) ] F,

F = 1.0 + 6S ,
2 0.5
[ 2
S = (x - 0.3100) + ( y - 0.3162) ] .

S is applicable for illuminant C. The need for such an adjustment was ascribed
to Takasaki’s chromatic crispening effect. For a combined set of data the
formula provided modest improvement over the FMC II or the unmodified
Glasser formula.
McLaren’s work with multiple linear regression had shown that it was
useful to consider separately three components of the total color difference,
assumed to be related in a euclidean manner: metric lightness, metric chroma,
and metric hue differences. Metric chroma was calculated as the square root
of the sum of the squares of the cartesian coordinates in the Adams-
Nickerson chromatic diagram and metric lightness and metric chroma were
subtracted in the euclidean manner from total color difference to result in
metric hue difference:
2 0.5
[ 2
DH = (DE ) - (DL) - (DC )
2
] . (6-21)

While the L, a, b view of the system is the cartesian coordinate view of the
euclidean space, that of L, C, H is the polar coordinate view.
Using new experimental data, and expanding on McLaren’s multiple linear
regression, R. McDonald in 1974 found that he could obtain significant
improvement in correlation between visual and calculated data by adjusting
the total color difference as a function of metric chroma as follows:

DE
DE a = , (6-22)
1 + 0.022C

where DEa is the equivalent color difference at the achromatic point and C is
the chroma value.

6.11 THE CIE 1976 L*a*b* AND L*u*v* SPACES

Because of the mentioned large number of different formulas used in

industry, the CIE organization in the early 1970s became increasingly aware
of the need to promote uniformity of practice in industrial color control.
However, the emerging picture was less than clear, and superiority had been
claimed for various formulas and in respect to frequently changing sets of
visual data. After much discussion a compromise was found, and in 1976 two
formulas were recommended for study and “in the interest of uniformity of
230 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

usage”(CIE,1976).One formula is a simplified version of the Adams-Nickerson

formula, abbreviated CIELAB, the other the earlier mentioned modified
version of MacAdam’s 1937 linear transformation of the CIE chromaticity
diagram, abbreviated CIELUV.The two formulas are defined as follows:

CIELUV
0.333
ÊYˆ
L* = 116 - 16,
Ë Yn ¯

u* = 13L * ( u' -u'n ),

v* = 13L * ( v' -v'n ),

2 0.5
[ 2
* = (DL *) + (Du *) + (Dv *)
DEuv
2
] ,

4X
u' = ,
X + 15Y + 3Z
9Y
v' = ,
X + 15Y + 3Z (6-23)

where

4Xn
un¢ = ,
X n + 15Yn + 3Z n
9Yn
vn¢ = ,
X n + 15Yn + 3Z n

and Xn, Yn, Zn are the tristimulus values of the nominally white object color
stimulus. The optimal object color solid and the spectral trace derived from
this formula are illustrated in Fig. 6-11.

CIELAB
0.333
Y
L* = 116Ê ˆ - 16,
Ë Yn ¯

È X ˆ 0.333 Ê Y ˆ 0.333 ˘
a* = 500 ÍÊ - ˙,
ÎË X n ¯ Ë Yn ¯ ˚

È Y 0.333 Ê Z ˆ 0.333 ˘
b* = 200 ÍÊ ˆ - ˙,
ÎË Yn ¯ Ë Zn ¯ ˚
2 0.5
[ 2
DEab * = (DL *) + (Da *) + (Db *)
2
] , (6-24)
 THE CIE 1976 L*a*b* AND L*u*v* SPACES 231

Fig. 6-11 Projective view of the L*, u*, v* object color space for the CIE 10° standard observer
and illuminant D65 (inner contour) and the spectrum locus. From Judd and Wyszecki (1975).

where Xn, Yn, Zn are the tristimulus values of the nominally white object color
stimulus. A slightly different calculation for L*, a*, and b* applies for low tris-
timulus value ratios. For X/Xn, Y/Yn, and Z/Zn > 0.01, instead of cube roots,
factors f are applied to the ratios, determined as follows: f(Y/Yn) = 7.787 (Y/Yn)
+ 16/116 and comparably for Y and Z. This adjustment is valid for the L*, a*,
and b* scales.
Differences can also be calculated in a polar coordinate version, where

Metric chroma C* = [(a*)2 + (b*)2]0.5

Hue angle hab = arctan (b*/a*)

From lightness and metric chroma differences, the total color difference DE is
calculated as follows:
2 0.5
[ 2 2
DE = (DL *) + (DC *) + (DH *) ] , (6-25)

where DH* = [(DE)2 - (DL*)2 - (DC*)2]0.5 and DC* = [(Da*)2 + (Db*)2]0.5. The
optimal object color solid and the spectral trace derived from this formula are
illustrated in Fig. 6-12.
As mentioned before, the former formula is primarily of interest to light-
ing engineers as it provides for additivity of light mixtures. The latter was rec-
ommended for use with object colors. CIELAB is a simplification of, but no
advancement over, the Adams-Nickerson formula, and it was apparent at the
232 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-12 Projective view of the L*, a*, b* object color space for the CIE 10° standard observer
and illuminant D65 (inner contour) and the spectrum locus. From Judd and Wyszecki (1975).

time of its recommendation that further modification was necessary to im-

prove the correlation of calculated with average visual color differences.

6.12 FRIELE’S FCM FORMULA

In 1978 Friele published a new formula optimized against various sets of visual
data. It was a modification of his earlier effort resulting in the FMC metric.
Friele had created psychometric scales for the various sets of acceptability data
that he used in addition to perceptibility data. The formula provided signifi-
cant improvements in correlation for these data, compared to CIELAB.

FCM

R = 0.760 X + 0.401Y - 0.124Z ,

G = -0.484 X + 1.381Y + 0.079Z ,

B = 0.847Z . (6-26)

For R > G,
13
Ê 1-Y ˆ
T = 125Y 1 3 .
Ë R1 3 ¯
 FRIELE’S FCM FORMULA 233

For R < G,
13
0.760 Ê 1-Y ˆ
T= 125Y 1 3 .
-0.484 Ë G1 3 ¯
2 -1 2
Y ÈÊ 0.085 x 4 3 ˆ 2˘
D=Ú ÍË + (0.055Y 2 3 ) ˙ dx. (6-27)
B
Î Y 2 3 ¯ ˚

L = 18Y 1 3 ,

a = T - 0.3D,

b = D - 0.3 T ,
1 2
Chroma = (a 2 + b 2 ) ,

0.024R 4 3
t= ,
Y2 3
1 2
ÈÊ 0.085B4 3 ˆ 2 ˘
d =Í + (0.055Y 2 3 )˙ .
ÎË Y ¯
2 3
˚

c 2 = T 2 + D2 ,
2
ÊD ˆ
a = arctan 2 ,
ËT ¯

f = 1.6[1 - exp(-0.0015c 2 )] sin 2a + exp(-0.0015c 2 ),

DL = 6Y -2 3 DY

Ê X ˆ Ê Z ˆ
tDT = 0.760 DX - DY - 0.124 DZ - DY ,
Ë Y ¯ Ë Y ¯

Ê Z ˆ
tDD = -0.847 DZ - DY ,
Ë Y ¯

2.5 0.5
DE =
1 + 0.01Y
[ 2 2
( fDL1 ) + (DT 2 ) + (DD) - fDTDD . ] (6-28)

Parameter f1 determines the weight of the lightness difference relative to the

chromatic difference. It ranges from 0.4 to 1.0 for different data sets.The model
represents an implementation of the Müller three-stage theory of color vision
mentioned earlier. R, G, and B represent the first, photopigment stage, and L,
T, and D the second stage opponent-response functions at the receptor level.
To get to the third, opponent stage at the “optic nerve” level, T and D are
234 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

modified, resulting in the rectangular coordinate system L, a, b. In this system,

somewhat comparable to the CIELAB space, the unit difference contours are
ellipsoids, rather than spheres, and the contours have to be normalized for the
effect of chromatic crispening and the ellipse shape of the unit contour with
the parameters t and d. Friele also implemented the controversial findings by
Schultze and Gall (1971) that unit small color difference ellipsoids are not
tilted toward the achromatic axis of color space.
The FCM formula, despite the improved correlation with visual data, did
not receive any significant practical use, perhaps because of its complexity and
the fact that its results were soon equaled by further optimization of the
CIELAB formula.

6.13 RICHTER’S LABHNU2 FORMULA

In the late 1970s K. Richter proposed a series of formulas as good models for
the Munsell and the OSA-UCS systems (Richter, 1980). Of these only the non-
linear LABHNU2 formula will be mentioned:
2 3
[(x y) + (1 6)]
a¢ = ,
15
13
[(z y) + (1 6)]
b¢ = , (6-29)
12

where x, y, z are CIE chromaticity coordinates. The CIELAB lightness formula

is used for lightness difference calculation

6.14 WEIGHTING OF METRIC LIGHTNESS, CHROMA,

AND HUE DIFFERENCES

JPC79 Formula
As mentioned above, in 1976 McLaren optimized weights for the three color
difference components based on the Adams-Nickerson formula. He found that
the optimum weights varied by set of visual data. McDonald continued to
pursue individual adjustment of the three color difference components and
developed new industrial visual data. In 1980 he proposed a formula employ-
ing continuous weight adjustment for all three metric components. It was
based on an extensive set of 640 samples around 55 standards distributed over
a significant portion of the object color solid. The samples consisted of dyed
spun polyester sewing thread and the visual evaluations (acceptability judg-
ments) were performed by eight industrial color matchers. An additional set
of some 8500 judgments against 600 color standards by a single observer in an
 WEIGHTING OF METRIC LIGHTNESS, CHROMA, AND HUE DIFFERENCES 235

industrial dyehouse was also included. McDonald’s color difference formula,

called the JPC79 formula, is based on the ANLAB formula as follows:
2 2 2 0.5
ÈÊ DL ˆ Ê DC ˆ Ê DH ˆ ˘
DEJPC 79 =Í + + ˙ , (6-30)
ÎË SL ¯ Ë SC ¯ Ë SH ¯ ˚

where DL, DC, and DH are, respectively, the lightness, chroma, and hue differ-
ences calculated from ANLAB,

S L = 0.08195L (1 *0.01765L),

SC = [ 0.0638C (1 + 0.0131C )] + 0.638,

S H = SCT ,

T = 1 if C = 0.38, otherwise

T = 0.38 + 0.4 cos(h + 35) , unless h is between 164∞ and 345∞, then

T = 0.56 + 0.2 cos(h + 168) .

The JPC79 formula represents the prototype of several further enhancements

fitting formulas with the help of various functions to ellipsoids optimized to
average visual data. It represents a practical, empirical approach to dealing
with the non-euclidean nature of a uniform color space.

CMC (l : c) Formula
This formula was slightly modified from JPC79 and based on CIELAB com-
ponent differences by the Color Measurement Committee of the Society of
Dyers and Colourists in England (Clark, 1984). It has been standardized in
England and in the United States, and is recommended by the International
Standards Organization (ISO). It is defined as follows:
2 2 2 0.5
ÈÊ DL* ˆ Ê DC* ˆ Ê DH* ˆ ˘
DECMC =Í + + ˙ , (6-31)
ÎË lSL ¯ Ë cSC ¯ Ë SH ¯ ˚

where
S L = 0.04097L* (1 + 0.01765L*) unless L* < 16, then S L = 0.511,

SC = [ 0.0638C* (1 + 0.0131C*) + 0.638,

S H = SC (TF + 1 - F ),
0.5
{
F = (C*)
4
[(C*) 4
+ 1900 ]} ,

T = 0.38 + 0.4 cos(h + 35) ,

236 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

and l and c are additional weights adjusting the relative weight of lightness
and chroma differences. Analysis has shown that for flat surface materials
(paints, plastics) viewed in sharp juxtaposition values of 1 in both cases are
appropriate, while for textile materials, the value l = 2 was found to improve
correlation. CMC found wide international usage in industry and continues to
be used in many firms.

6.15 NEW SETS OF VISUAL DATA

In 1978 the CIE issued a set of guidelines for coordinated research on color
difference equations (Robertson, 1978). In these guidelines particular empha-
sis was placed on five color centers: gray, yellow, red, blue, and green. As a
result several experimenters provided data for these centers but new data for
many other centers were also established.

Luo and Rigg

As a prelude to their formula fitting (see below) R. Luo and B. Rigg assem-
bled 13 previously published sets of perceptibility and acceptability data (Luo
and Rigg, 1986). They fitted separate ellipsoids in x, y, Y space to each subset
and compared these. The main difference for ellipsoids in the same neighbor-
hood of global color space appeared to be a size factor (see Fig. 6-13 showing
chromatic ellipses). This is understandable as the implicit acceptability toler-
ances of different observer groups are likely to be different and different
scaling techniques had been used in the perceptibility judgments. The relative
sizes were adjusted by judging 400 pairs of samples representing 70 color
centers against a gray scale. Luo and Rigg also deleted certain subsets that
they found to be internally inconsistent and, as a result of additional experi-
ments, modified certain ellipses. No consideration was given to differences in
surround in various data sets. The result was 132 ellipses that were considered
to be reliable and formed a reasonably regular pattern in the CIE chromatic-
ity diagram. Luo and Rigg found little difference between perceptibility and
acceptability ellipses. The data are available on the University of Derby Web
site (Derby, 1999).

Cheung and Rigg

In support of the CIE effort M. Cheung and Rigg established in 1986
suprathreshold small color difference data for the suggested five centers.
The samples consisted of dyed wool fabric, with 59 to 82 pairs per center,
assessed by 20 observers against a standard difference pair under artificial day-
light and tungsten light against a neutral gray background of Y = 13. The
samples had CIELAB color differences from 1 to 9 units from the respective
standards.
 NEW SETS OF VISUAL DATA 237

Fig. 6-13 Ellipses fitted to unadjusted data of the Luo and Rigg data set, enlarged 1.5 times,
in a portion of the CIE chromaticity diagram. From Luo and Rigg (1986).

Badu and Rigg Large Difference Data

In connection with a Ph.D. thesis, A. Badu prepared in 1986 a set of data with
relatively large differences. Samples were prepared to form a grid of approx-
imately equilateral triangles on the a*, b* diagram. Some samples also
involved lightness differences. There were 408 sample pairs, 238 consisting of
nylon dyeings and 170 of glossy paint samples. They were visually evaluated
by a group of 20 observers using a gray scale as standard.

Witt (1987–1990)
In 1987 K. Witt reported on the results of a threshold experiment using 50 to
64 sample pairs in four of the five CIE color regions prepared with high gloss
acrylic paint. Some 24 observers determined at least four times (some
observers 10 times) under simulated daylight and against a surround of Y =
20 if they could perceive a difference between the pairs. Ellipses were calcu-
lated from the resulting thresholds for individual observers and for sets of 4
to 22 observers with from 22 to 118 repetitions. Considerable intra- and inter-
238 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

observer variability was noted. Average color differences of the thresholds

were approximately 0.4 CMC (2 : 1) units.
In 1990 Witt reported a further experiment involving the five CIE color
centers. In addition to determining thresholds, observers were asked to assess
perceptible differences in terms of hue, chroma, and lightness. Another
purpose of the experiment was to determine the effects on the thresholds of
changing the surround lightness and of a gap between the samples. Interest-
ingly better agreement was found between the chromatic ellipse axis and the
hue judgment axis in the CIE chromaticity diagram than in the a*, b* diagram.
A small surround lightness effect and a larger gap effect were found.

RIT-DuPont Data
In an effort to develop a set of high-reliability small color difference percep-
tibility data D. Alman of DuPont and Berns of RIT (Rochester Institute of
Technology) collaborated in 1990 and published the RIT-DuPont data (Berns
et al., 1991). They are based on 156 acrylic lacquer spray-painted sample pairs
on primed aluminum panels representing 19 color centers. The sample pairs
did not have a common reference and were arranged so as to represent spe-
cific vector directions in CIELAB color space. The sample pairs, presented
edge-to-edge on unprimed aluminum panels, were evaluated by 50 observers
against a near gray standard pair presented identically. Probit analysis was
used to establish psychometric scales, and the colorimetric values were set so
that visual differences equal to that caused by a 1.0 DE CIELAB lightness dif-
ference resulted. Ellipsoids in CIELAB space have later been optimized to
the visual data. The ellipses in the a*, b* plane show fair agreement with those
of Luo and Rigg (see Fig. 5-29). Both show ellipses more or less pointed toward
the origin of the diagram, except for blue colors. The data have been published
and are also available on the University of Derby Web site.

Leeds Data
Two sets of data were established by D. H. Kim and J. H. Nobbs at Leeds Uni-
versity in 1997. They consist of matte painted samples. Sample set PC consists
of 152 pairs evaluated juxtaposed by 15 observers against a near neutral ref-
erence pair. Sample set GS consists of 204 sample pairs evaluated against a
gray scale by 12 observers. Color differences ranged from 0.4 to 3.7 CIELAB
units in the two sets. The data are available on the University of Derby Web
site.

Pointer and Attridge Large Difference Data

In 1997 Pointer and Attridge published the results of evaluation using a set of
samples with relatively large differences (from approximately 2 to 20 CIELAB
units). The samples were prepared by varying red, green, and blue exposures
 NEW FORMULAS 239

on photographic paper. Samples were mounted in gray, commercial 35 mm

slide mounts, and displayed against an identically mounted near neutral stan-
dard pair. There are four subsets (value 5, saturated, light, and dark) each with
six to eight colors in the same categories (red, green blue, cyan, magenta,
yellow, skin, and neutral). The total set consists of 384 pairs. Comparison of
pairs of samples and standard were made by nine observers against a neutral
pair with a designated visual color difference of 10 units. Sixty percent of the
sample pairs have less than 6 units of CIE94 color differences; that is, the set
contains a large component of small color differences. The data are available
from the authors.

Witt (1999)
In 1999 Witt published a set of data involving the five CIE standard color
regions. For each region there are approximately 30 painted samples. These
have been evaluated by from 10 to 13 observers in approximately 85 different
combinations of two each, thus providing a detailed evaluation in each region
of color difference vectors in two and three dimensions. Comparisons were
made against a specially prepared gray scale (near logarithmic) found to be
visually uniform, with the result expressed in steps and third or quarter frac-
tion steps of the reference scale. Color differences ranged from approximately
0.3 to 8 CIELAB units. Means and standard deviations of the visual judgments
were calculated. The coefficient of variation of the visual results ranged from
15% to 60%. The data have been published in Witt’s paper and are available
on the University of Derby Web site.

Guan and Luo Large Difference Data

These authors published in 1999 results of evaluations of a set of relatively
large differences (6 to 21 CIELAB units). The samples form a grid of approx-
imate squares in the a*, b* diagram at three different levels of L* (40, 50, and
60). Samples were compared by ten observers not only in the a*, b* directions
but also on diagonals, involving changes in both a* and b*. The samples consist
of wool dyeings and they were evaluated in 292 pairs against an eight-step gray
scale, also consisting of wool dyeings.

6.16 NEW FORMULAS

BFD (l:c)
Using the above-described composite data set, Luo and Rigg (1987) optimized
a formula within the general CMC framework as follows:
2 2 2 0.5
ÈÊ DL ˆ Ê DC* ˆ Ê DH* ˆ Ê DC* ˆ Ê DH* ˆ ˘
DEBFD =Í + + + RT , (6-32)
ÎË l ¯ Ë cDC
¯ Ë DH
¯ Ë DC ¯ Ë DH ¯ ˙˚
240 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

where

DC = 0.035C * (1 + 0.0365C *) + 0.521,

DH = DC (GT ¢ + 1 - G),
0.5
{
G = (C *)
4
[(C ) 4
+ 14000 ]} ,

T ¢ = 0.627 + 0.055 cos(h - 254) - 0.040 cos( 2 h - 136∞) + 0.070 cos( 3h - 32 ∞ ),

+ 0.049 cos( 4 h + 114 ∞ ) - 0.015 cos( 5h - 103 ∞),

RT = RH RC ,

RH = -0.260 cos( h - 308 ∞) - 0.379 cos( 2 h - 160 ∞) - 0.636 cos( 3h + 254 ∞ ),

+ 0.226 cos( 4 h + 140 ∞) - 0.194 cos( 5h - 280 ∞),
0.5
{
RC = (C *)
6
[(C ) ] + 70000}
6
,

L = 54.6 log10 (Y + 1.5) - 9.6.

DC* and DH* are CIELAB chroma and hue differences. The overbar indicates
the mean value between standard and sample. Constants l and c are compa-
rable to the corresponding constants in the CMC equation.
While DC is a slight modification of SC in CMC DH is a complex function
that corrects for experimental variation in the relationship between hue angle
difference and visual hue difference around the hue circle. An additional term
is added to accommodate ellipses of bluish colors not directed toward the
origin.

SVF Formula
Also in 1986, a different approach was taken by T. Seim and A. Valberg.
They fitted a formula to the Munsell Renotation data based on cone sensi-
tivities. In the first step tristimulus values are converted to cone sensitivities
that are then centered on white (S1 comparable to L, S2 comparable to M and
S3 comparable to S). The lightness value V is obtained with a hyperbolic
formula that represents a close fit of the polynomial Munsell value formula:

VY = 40v1 (Y ), (6-33)

where
0.51
(Y - 0.43)
v1 (Y ) = 0.51
.
(Y - 0.43) + 31.75
 NEW FORMULAS 241

Following Adams, the same model is applied to the two chromatic dimensions.
Opponent color signals are calculated in two steps:
p1 = v1 (S1 ) - v1 (Y ),

p2 = v1 (Y ) - v1 (S3 ) if S3 £ Y ,

p2 = v2 (Y ) - v2 (S3 ) if S3 > Y (6-34)

where
0.86
[(Y k(VY )) - 0.1]
v2 (Y ) = 0.86
,
[(Y k(VY )) - 0.1] + 103.2
k(VY ) = 0.140 + 0.175VY .
The final opponent-color functions are calculated as

F1 = 700 p1 - 54 p2 ,

F2 = 96.5 p2 .
Color differences are calculated as
2 0.5
[ 2 2
DESVF = (DF1 ) + (DF2 ) + 2.3(DVY ) ] .

Unlike in the Müller-Judd approach the p level does not involve the
subtraction of cone responses but the subtraction of Y from the response of
L in one case and the response of S in the other. As a result the yellow-blue
axis is established in the final form (except for scaling) at the p level. The
green-red axis needs to be rotated; thus semicolon the subtractive form of F1.
A different hyperbolic function is applied to S > Y to account for the appar-
ent change in the modulation in the Munsell system of the S function of yel-
lowish and bluish colors, which we will encounter again later.
The formula provides a reasonably good fit to the Munsell system even
though the yellowish and bluish constant hue lines are significantly curved.
The formula was also applied to the OSA-UCS data where it is less success-
ful because the implicit modulations in this system are different from those
implicit in the Munsell system. Seim and Valberg applied the formula also to
small suprathreshold color difference data with reportedly good results.

CIE94
In the early 1990s CIE technical committee 1–29 investigated three data sets
considered reliable in regard to the relationship between DL* and L*, DC*, and
C*, and DH* and hue angle h (Berns, 1993). Considerable scatter was found in
all three cases.As a result the committee optimized simple weights for CIELAB
lightness, chroma, and hue differences and issued the formula known as CIE94:
242 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

2 2 2 0.5
ÈÊ DL * ˆ Ê DC * ˆ Ê DH * ˆ ˘
DE *94 = Í + + ˙ , (6-35)
ÎË kL SL ¯ Ë kC SC ¯ Ë kh Sh ¯ ˚
where
SL = 1,

SC = 1 + 0.045C *,

SH = 1 + 0.015C * .

The k constants are additional weights on the lightness, chroma, and hue dif-
ferences. Under reference conditions they all equal 1. For textile applications
kL typically has a value of 2. No weighting of lightness differences was made
because the scatter in the combined experimental data (without consideration
of surround lightness) was too large for a statistically meaningful weight.

Kim and Nobbs Weights for CIELAB

In 1997 Kim and Nobbs proposed new lightness, chroma, and hue difference
weights for the CIELAB formula based on an analysis of five data sets includ-
ing the Luo and Rigg, RIT-DuPont, and Leeds data:
2 2 2 0.5
È ( DL* SL ) (DC * SC ) + (DH * SH ) + SR DC * DH * ˘
DELCD =Í 2
+ 2 ˙ , (6-36)
Î KL KCH ˚
where
2
S L = 1 - 0.01L* + 0.0002(L*) , if L* < 50, then S L = 1.0,

SC = (1 + 0.045C *)SCH ,

S H = (1 + 0.015C *)S HH ,

[ 3
]
S R = -C * (2 + 0.07C ) sin(2DQ),

{
DQ = 30 exp -[ h - 275 25] ,
2
}
K L = KCH = 1.0 for nontextile samples, K L = 1.5 for textile samples.

Complex sinusoidal SCH and SHH functions have been developed that improved
the correlation between calculated and visual data for some data sets but not
others:
SCH = 1 + 0.07 sin(h) - 0.16 cos(2 h + 250) - 0.05 cos(3h) - 0.03 cos(4 h),

S HH = 1 + 0.03 cos(h + 60) + 0.12 cos(2 h) + 0.12 cos(3h) - 0.07 cos(4 h - 45).
 NEW FORMULAS 243

Integrating Weights
Proposals to integrate the weights on chroma and hue differences in CIE94
directly into the calculation of modified a* and b* values were made by E.
Rohner and D. C. Rich in 1996, by H. G. Völz in 1998 and 1999, and by K.
Thomsen in 2000 (see also DIN99 below). In Thomsen’s version,
a * ¢ = a *f (C *),
b * ¢ = b *f (C *), (6-37)
where
ln(1 + 0.0531301C *)
f (C *) = .
0.0500951C *
The two numerical factors have been optimized against CIE94 from 200,000
positions at constant L* in the a*, b* diagram. The maximum discrepancy
found by this method, compared to conventional calculation using CIE94, is
10.5%. Given the variability of the visual data behind the SX weights, this is
not problematical. The only direct advantage of such an approach is that
euclidean relations are maintained. However, the additional optimization
parameters added in CIEDE2000 (see below) would require new, more
complex integration efforts to maintain euclidean relationship with question-
able ultimate value, as all are based on the uncertain fundament of CIELAB.
In addition CIEDE2000 is only applicable from threshold to 6 to 8 units of
total difference.

Guan and Luo Large Difference Formula

In 1999 Guan and Luo fitted a formula to their large difference data men-
tioned above and to other large difference data sets. The formula, named
GLAB by the authors, follows the CIE94 format with the following weights:
SL = 0.76, SC = 1 + 0.016 Csb , SH = 1.0,
where Csb = (C*Std C*Spl)0.5; that is, the chroma value used is the average of those
of the standard and the sample being compared. The factors KL, KC, KH were
taken as 1. From this proposal it is evident that different formulas and/or
different modification functions are required for small and large differences.

Lübbe Adjustment for Surround Lightness

In 1999 Lübbe proposed a formula for adjusting the L* scale for the effects
of surround
L * ¢ = L * - f (L *u , L *)(L *u -L *)
where L*¢ is the adjusted lightness value, L* the original value, and L*u the
original L* value of the surround. f is an experimentally determined factor
244 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

depending on the lightness of the test and the surround fields. It results in S-
shaped functions of the relationship between L* and L*¢.

DIN99 Formula
A new color space and difference formula was developed in 1999 in Germany
(DIN 6176 2000). Its purpose is to apply the integration method proposed by
Rohner and Rich (1996) to produce an euclidean space and difference formula
with good performance against various sets of small color difference data. The
formula went through several stages of development, and presently there are
four versions in existence. Only the latest (and best performing) formula
DIN99d is given here. It is based on the CIELAB formula and the X tristim-
ulus values of reference, and test colors have been adjusted by the procedure
proposed by Kuehni (1999).

X ¢ = 1.12 X - 0.12Z ,
L99d = 325.22 ln(1 + 0.0036L *),
e = a * cos(50 ∞) + b * sin(50 ∞),
f = 1.14[-a * sin(50 ∞) + b * cos(50 ∞)],
0.5
G = (e 2 + f 2 ) ,
C99d = 22.5 ln(1 + 0.06G),
Ê fˆ
h99d = arctan + 50 ∞,
Ë e¯
a99d = C99d cos(h99d ),
b99d = C99d sin(h99d ),
1 0.5
DE99d = (DL299d + Da992 d + Db992d ) . (6-38)
kE

The formula has been tested against the same data used in testing CIEDE2000
(see below) and performs somewhat better than CMC and CIE94 but
slightly inferior to CIEDE2000. Its error against the combined test data set
is 35% (compared to 38% for CIE94 and 33% for CIEDE2000) (Cui et al.,
2002).

Kuehni Optimization of the CIE94 Formula

In 2001 Kuehni published a report on his analysis of the RIT-DuPont, Witt,
Leeds GS and PC, as well as the Pointer-Attridge large color-difference data
(Kuehni, 2001b). As a result of the analysis, and limiting changes to those
applicable to all visual data sets individually, he recommended a modi-
fied CIE94 formula. The first step consists of an adjustment of the x color-
matching function as follows:
 NEW FORMULAS 245

xmod = 1.1x - 0.1z . (6-39)

The factor 1.1 is applicable to the CIE 10° observer data. It has a value of 1.06
for the 2° observer data. This adjustment results in a rotation of the b* axis
that better aligns the unit ellipses of blue colors with lines of constant hue
angle.
The lightness scale is revised to make it dependent on surround lightness,
resulting in a true opponent color scale. An SC type adjustment for the light-
ness crispening effect is introduced:

0.333 0.333
ÈÊ Y ˆ Ê Ys ˆ ˘
LŸ = 116 Í - ˙,
ÎË Y0 ¯ Ë Y0 ¯ ˚
DLŸ
DL = , (6-40)
SL

where YS is the luminous reflectance of the surround and Y0 that of the

illuminant,

SL = 1 + 0.010LŸ .

These changes resulted in relative improvements in prediction from 5% to

20% depending on the data set. Other changes were found to result in
improved correlation with visual data for individual sets but not for all sets.
Surprisingly, when used as a last step in the optimization, hue angle dependent
functions to adjust the size of the hue difference component such as investi-
gated by Kim and Nobbs (1997) or the function used in CIEDE2000 (see
below), were found to have no meaningful positive effect on the correlation
for any of the data sets investigated.
Kuehni found that the introduction of nonsystematic size adjustment
factors for individual color groups, ranging from approximately 0.75 to 1.5,
improved correlation significantly in all cases. This indicates considerable vari-
ation among and within data sets in how observer groups judge the size of
color differences in specific locations in color space, compared to the predic-
tion by the formula. The source of this variation has not yet been systemati-
cally investigated.

CIEDE2000
In 2001 Luo, Cui, and Rigg, based on extensive analysis of several sets of per-
ceptual color difference data, proposed a modified BFD formula they initially
called M2b. It is based on CIELAB with the following analytically arrived
modifications:
246 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

L¢ = L*,

a¢ = (1 + G)a*,

b¢ = b*,
2 0.5
[
¢ = (a¢) + (b¢)
Cab
2
] ,

Ê b¢ ˆ
¢ = tan -1
hab ,
Ë a¢ ¯
0.5
Ï È 7 ¸
Ô Í
G = 0.5Ì1 -
(Cab* ) ˘
˙ Ô
(6-41)
7 ˝,
Ô ÍÎ (Cab* ) + (25) ˙˚
7
Ô
Ó ˛

where C *ab is the arithmetic mean of the C*ab values for a pair of samples.

DL ¢ = Lb¢ - L¢s ,

¢ = C ab
DC ab ¢ ,b - C ab
¢ ,s ,

È 0.5 ¢ ˆ˘
Ê Dhab
¢ = Í2 C ab
DH ab (
¢ ,bC ab
¢ ,s ) sinÁ
Ë 2 ¯˚
˜ ˙,
Î

¢ = hab
where Dhab ¢ ,b - hab
¢ ,s .

The subscripts b and s refer to the comparison sample and the standard
sample, respectively.

2 2 0.5
ÈÊ DL¢ ˆ 2 Ê DCab ¢ ˆ
¢ ˆ Ê DH ab Ê DCab ¢ ˆ˘
¢ ˆ Ê DH ab
DE = Í +Á ˜ +Á ˜ + RT Á ˜ Á ˜˙ , (6-42)
ÍÎË kL SL ¯ Ë kC SC ¯ Ë kH SH ¯ Ë kC SC ¯ Ë kH SH ¯ ˙˚

where
2
0.015(L ¢ - 50)
SL = 1 + ,
2 0.5
[ 20 + (L ¢ - 50) ]
SC = 1 + 0.045Cab¢ ,

SH = 1 + 0.015Cab¢ T ,

with T = 1 - 0.17 cos(h a¢b - 30°) + 0.24 cos(2h¢ab) + 0.32 cos(3h a¢b + 6°) - 0.20
cos(4h a¢b - 63°). RT = -sin(2DQ)RC and
 NEW FORMULAS 247

2
Ô
DQ = 30 expÌ- Í
( )
Ï È hab¢ - 275 ∞ ˘ ¸
˙ ˝,
Ô
ÔÓ ÍÎ 25 ˙˚ Ô˛
7 0.5
Ê Cab¢ ˆ
RC = 2Á 7 ˜ .
Ë Cab¢ + 25 7 ¯

L¢, C ab
¢ , and h ab
¢ are the arithmetic means for a pair of samples of the respec-
tive individual values.
This formulation corrects for the slanted ellipses near the negative b*
axis, contracts the a* axis near the neutral point so that ellipses located in
that area can be treated as such, introduces a new lightness weighting func-
tion, and uses the hue difference weighting function T proposed by Berns
at the Warsaw CIE meeting in 1999 (Berns, 2000). The new lightness dif-
ference weight is adjusted, without explicitly stating so, for a surround with
L* = 50. For the combined set of all data the new formula has an error of
prediction of 33% compared to 38% for CIE 94 and CMC. The formula per-
forms about equally well as CIE 94 for the RIT-DuPont data set and inferior
to BFD for the Luo-Rigg (now BFD) set. It is marginally better with the Witt
set and distinctly better with the Leeds and a new set of data, BIT, involving
CRT display colors. The formula was adopted as CIEDE2000. Unit difference
ellipses in the a*, b* diagram generated by this formula are illustrated in
Fig. 6-14.

Sections 6.5 to 6.16 have focused on the development of industrially impor-

tant color difference formulas for use in color quality control. This develop-
ment has proceeded largely independent of development of models of color
vision and their implications for color spaces. Such efforts in a few cases go
much beyond those of industrial color difference formulas in that they attempt
to encompass all aspects of color vision while color difference formulas tend
to only reflect one set of test conditions: daylight illumination and a single,
achromatic surround. Visual data sets involving physical samples are expen-
sive to prepare, and visual evaluations of existing ones require different
formula optimizations, for reasons that are not clear. Less expensive color dif-
ference evaluation using monitor colors is in its infancy, and the monitor con-
ditions that produce identical results by the same observers to those obtained
with comparable physical samples in simplified conditions are not yet known.
In the area of formula fitting the CIE recommendation of the CIELAB
formula had a strong effect in that since then most formulas, for better or
worse, have been based on the foundation of CIELAB. As will be discussed
in Chapter 9, progress over what is available now with CIEDE2000 depends
on new systematic visual data and on a color space model more in agreement
with facts than CIELAB.
248 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-14 Unit ellipses in the CIELAB a*, b* diagram as calculated from the CIEDE2000
formula. From Luo, Cui, and Rigg (2001).

6.17 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS

OF COLOR VISION

Since the 1950s several comprehensive models of color vision have been pro-
posed, some of which have had an impact on mathematical formulations of
color space and color differences. A detailed discussion of this subject is
outside the scope of this text and only a brief overview will be presented.

Müller-Judd
G. E. Müller (1850–1934) was a German psychophysicist much interested in
color vision. He adopted Hering’s theory of three reversible photochemical
substances and attempted to bring an opponent color model in line with
psychological facts. Müller developed the concept of cortical gray as repre-
senting the neutral state of color vision. In a series of articles published in
1896–97 titled Zur Psychophysik der Gesichtsempfindungen (On the psy-
chophysics of the visual sense) he described a three-stage process of color
vision and later expanded on the subject in a book-length treatise Über die
 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS OF COLOR VISION 249

Farbenempfindungen (On color sensations, 1930). The first stage represents

cone absorptions, followed by an intermediate differential second stage of
coding and final neural coding in the third stage, resulting in chromatic and
achromatic opponent color signals.The first two stages of the process are today
loosely supported by neurophysiology, while the third stage continues to
remain a hypothesis.
Beginning in 1949 and continuing until the end of his life, Judd worked on
a model of the Müller theory, attempting to express it in terms of CIE tris-
timulus values. Three cone responses are derived from tristimulus values as
follows:
P1 = 3.1956X + 2.4478Y - 0.6434Z ,

P2 = -2.5455X + 7.0492Y + 0.4963Z ,

P3 = 5.0000Z . (6-43)

In the second stage the cone signals are converted to intermediate opponent
color signals as follows:

a1 = P1 - P2 ,

a 2 = 0.015P1 + 0.3849P2 - 0.4000 P3 . (6-44)

In the third stage these signals are converted to the final chromatic opponent
color signals as follows:
b1 = a1 - 0.6265a 2 ,

b 2 = a 2 + 0.1622a1 . (6-45)

The achromatic final signal is equal to the CIE tristimulus value Y. The chro-
matic signal calculation can be reduced to expressions of tristimulus values
with the following result:

b1 = 6.325(X - Y ),

b 2 = 2.004(Y - Z ),

b3 = Y . (6-46)

At the third stage this model bears similarity to the Adams zone theory model
and the Jameson-Hurvich model to be presented below. Judd used his imple-
mentation of the Müller theory to predict results of color vision impairment
and good agreement with wavelength discrimination data of protanopes, tri-
tanopes and normal dichromats was obtained (Judd and Yonemura, 1970). The
efforts by Friele to use the Müller framework to develop color difference for-
mulas were discussed above. The Müller framework has also been used by
S. L. Guth (see section below).
250 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Adams
As mentioned earlier, Adams offered a zone theory of color vision in 1923 and
a chromatic value diagram based on it in 1942. He equated cone sensitivity
with tristimulus values and applied the Munsell value function, representing
the modulation of the output of the “cones,” to all three tristimulus values. The
chromatic signals were calculated as follows:

a = 1.9(VX - VY ),

b = 0.72(VZ - VY ), (6-47)

where VX, VY, and VZ are square root Munsell value type functions of the CIE
tristimulus values X, Y, and Z that Adams equated with cone responses. The
constants were arrived at by comparing the ranges of chromatic value result-
ing from the color-matching functions and by bringing them into agreement
with the Munsell value scale. The ratio is 2.64, larger than the theoretical value
for a balanced linear system and larger than the value of 2.5 of the CIELAB
space. As seen earlier, this model was used in modified form as a basis for the
Adams-Nickerson color space and difference formula and the various formu-
las derived from it.

Hurvich and Jameson

In the early 1950s, as mentioned, Hurvich and Jameson developed an interest
in the Hering theory by then widely disregarded by color science (except in
Germany). They experimentally determined chromatic response functions by
evaluating the amounts of certain chromatic stimuli required to cancel the hue
of other stimuli. A typical result is shown in Fig. 6-15 and can be seen as in
agreement with the Hering theory. The two spectral functions can be inter-
preted to represent a greenness-redness system (filled circles) and a yellow-
ness-blueness system (open circles). Hurvich and Jameson took these
functions to be linearly related to CIE color-matching functions, and their
functions for white adaptation for the 1931 standard observer were shown in
the previous chapter (Fig. 5-13). The general validity of such functions rests
on the assumption that they are additive. Tests by Larimer and co-workers
(1974–5) indicated the greenness-redness system to be additive and the yel-
lowness-blueness system approximately so. The chromatic functions of Fig.
5-13 are approximated by the equations
a = x - y,

b = 0.4 y - 0.4 z , (6-48)

where x, y, z, are the CIE 1931 color-matching functions. These functions can
be expressed in relative terms as hue coefficients (Fig. 6-16), and the latter
 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS OF COLOR VISION 251

Fig. 6-15 Spectral opponent color (and lightness) functions determined from hue cancellation
experiments, of a single observer. From Hurvich (1981).

were found to be in good agreement with the results of spectral hue naming
experiments. Except for the weights, they are identical to the final Müller-Judd
functions and in basic agreement with the Adams functions. Jameson also
expressed the chromatic functions in terms of König type cone fundamentals
(1972).The Hurvich-Jameson functions provide good support, derived through
the paradigm of hue cancellation, for the basic ideas of Hering, Müller-Judd,
and Adams.
Hurvich and Jameson proposed a polar chromatic diagram they termed a
psychological diagram. It is derived by multiplying a spectral power distribu-
tion with their two chromatic functions a and b, based on their own experi-
mental data. The two functions form the axes of a polar diagram (see Fig. 6-17)
where saturation is calculated as the euclidean sum of the two chromatic
responses. Conceptually unique hues fall on the axes of this diagram. However,
their chromatic functions resemble the functions of equation (6-48), and
average experimental unique red and green have significant positive b values.
The diagram is essentially identical to a linear opponent diagram in polar coor-
dinate form based on CIE color-matching functions. For a description of the
work of Jameson and Hurvich related to the Hering system, see Hurvich
(1981).
252 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-16 Spectral hue coefficient functions for an average observer (Hurvich, 1981). Compare
with Figs. 4-8 and 4-9.

Fig. 6-17 The polar hue/saturation diagram of Hurvich and Jameson (Hurvich, 1981). Spec-
tral colors fall on the heavy line in the interior.
 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS OF COLOR VISION 253

Guth ATD Color Vision Model

A comprehensive color vision model has been developed by Guth and co-
workers since the early 1970s (Guth, 1972, 1991). It is described as a modern-
ized form of the Müller model. It was revised several times since the original
proposal, and the most recent version is known as ATD 01 (Guth, 2001). Since
different sets of visual data require different parameters and parameter func-
tions for best fit, there is no single version applicable to all aspects of color
vision. Guth does not consider ATD to be a model but “a quantitative theory
of color vision.”
CIE tristimulus values of the test and the surround are converted to mod-
ulated cone responses according to

0.70
L = (0.16 X + 0.56Y - 0.034Z ) ,
0.70
M = (-0.40 X + 1.16Y + 0.084Z ) ,
0.70
Z = (0.017Y + 0.27Z ) . (6-49)

Output from the three cone types is subjected to gain control by multiplying
them with their attenuation factors:

s
,
s + L + k(La - L)
s
,
s + M + k ( Ma - M )
s
, (6-50)
s + S + k(Sa - S)

where s = 200 and k = 5.5, and subscript a denotes the surround (adapting
field). The equations apply only if the cone response of the surround is larger
than that of the test field; otherwise, the difference between the test and sur-
round cone responses in the calculation of the attenuation factors is taken
as zero. The resulting cone responses after gain control (subscript g) are used
to calculate uncompressed responses for stage 1 and stage 2 mechanisms as
follows:

A1i = 3.75Lg + 2.64 Mg , A2i = 0.09 A1i ,

T1i = 6.90Lg - 6.90 Mg , T2i = 0.60T1i + 0.75D1i ,

D1i = -0.83Lg + 1.00Sg , D2i = - D1i , (6-51)

254 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-18 Spectral functions of Guth’s A, T, D color space.

where subscripts 1 and 2 refer to the mechanism stages and subscript i to the
initial uncompressed response. The uncompressed stage 2 opponent color
responses are illustrated in Fig. 6-18 for comparison to the Jameson-Hurvich
functions and other functions below.
In the next step the final compression for the second-stage ATD values are
calculated as follows (the compressed values for the stage 1 ATD values are
calculated in the same manner):

A2 = A2 i (200 + A2 i ),
T2 = T2 i (200 + T2 i ),
D2 = D2 i (200 + D2 i ). (6-52)

At stage 1 the brightness signal is equal to the vector sum of the compressed
achromatic and two chromatic components. Lightness is calculated as the
brightness compared to the brightness of the reference white (scale 100).
Colorfulness, chroma, and saturation are all calculated according to
0.50
Co = C = Sa = (T22 + D22 ) A2 .

Hue H is expressed as hue angle and calculated as H = arctan (D2/T2).

Color differences are calculated differently, according to their magnitude.
Small color differences (MacAdam ellipses, small suprathreshold differences)
 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS OF COLOR VISION 255

are calculated as the square root of the sum of the squares of the three com-
ponent vector differences at the compressed stage 1 level, while large color
differences are calculated in the same manner at the compressed stage 2 level.
Many kinds of data have been predicted, with varying success, with differ-
ent versions of the model. It has not found industrial use as a uniform color
space formula. As a color appearance model the ATD95 version has received
faint praise by Fairchild (1998).

OSA-UCS Space Formula

In connection with the development of the OSA-UCS system (see Chapters
2 and 8) the committee also developed a formula to describe the system in
terms of the CIE 10° observer data and illuminant D65 (MacAdam, 1974).
Tristimulus values are converted to cone sensitivity values that are much
different from those of Smith and Pokorny:

R10 = 0.799 X 10 + 0.4194Y10 - 0.1648Z10 ,

G10 = -0.4493 X 10 + 1.3265Y10 + 0.0927 X 10 ,

B10 = -0.1149 X 10 + 0.3394Y10 + 0.717Z10 . (6-54)

Two chromaticness and a lightness coordinate are calculated as follows:

g = C (-13.7R1 3 + 17.7G1 3 - 4 B1 3 ),

j = C (1.7R1 3 + 8G1 3 - 9.7 B1 3 ),

È Y01 3 - 2 ˘
L = 5.9 Í 13 ˙
,
ÍÎ 3 + 0.042(Y0 - 30) ˙˚

where
13
1 + 0.042(Y0 - 30)
C= ,
Y01 3 - 2 3

Y0 = Y (4.4934 x 2 + 4.3034 y 2 - 4.276 xy - 1.3744 x - 2.5643 y + 1.8103).

Variable C adjusts chromaticness for the lightness crispening effect by a mod-

ified Semmelroth (1970) formula. Variable Y0 adjusts lightness and chromatic-
ness for the Helmholtz-Kohlrausch effect with a formula modified from that
of Sanders and Wyszecki (1958). (Both factors are discussed in Chapter 5.)
The committee explicitly stated that the formula is not to be used for color
difference calculation of small color differences. The spectral R, G, B and g, j
functions of the space are illustrated in Fig. 7-17a and b.
256 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Cohen’s Fundamental Color Space

Beginning in 1982 J. B. Cohen and W. E. Kappauf began to describe what
Cohen later termed fundamental color space (FCS; Cohen and Kappauf, 1982;
Cohen, 2001). They found that all metamers of a particular spectral power
distribution (SPD) can be shown to consist of a common component (the
fundamental) and a variable component (metameric black), as predicted by
Wyszecki in 1953. Mathematically the fundamentals can be extracted from an
SPD with the help of an orthogonal projector Cohen called matrix R. Matrix
R projects spectral power distributions into the FCS, the mathematical space
of all possible fundamentals. The axes of FCS can be arbitrarily selected, but
Cohen proposed two preferred reference frames, the canonical frame where
one of the orthonormal functions is the equal energy function and the frame
where the three axes are orthogonal spectral vectors (455, 513, and 584 nm).
Since the CIE luminous reflectance function is different from the equal energy
and any of the orthogonal vector functions, neither configuration is com-
parable to the CIE X, Y, Z space.
There is no reason why FCS should be perceptually uniform, and it is not.
Burns and co-workers (1990) projected the spectral power distributions of the
color chips of the 1929 Munsell Book of Color as viewed under illuminant C
into two different versions of FCS. Only when using the luminosity function
as one of the three axes of FCS did the samples arrange themselves into layers
that correspond to Munsell value (Fig. 6-19). When the space is viewed from
the top (along the luminosity axis), they do not form concentric circles as one
might expect but a somewhat elliptical cloud. Data plotting confirms the ellip-
tical nature of the constant chroma contours. The spectral orthonormal F func-
tions for this situation are illustrated in Fig. 6-20. The two chromatic functions
are significantly different from opponent color functions such as shown in
Fig. 6-15.

The Cardinal Planes Space of Derrington, Krauskopf, and Lennie

As mentioned in Chapter 5, in 1984 Derrington and co-workers reported on
their findings regarding chromatic mechnisms in the lateral geniculate nuclei
(LGN) of macaque monkeys. Given the close genetic relationship to humans
results from macaques are considered relevant to human color vision. Mea-
suring parvocellular activity in the LGN of their test objects the authors iden-
tified two types of cells with opponent chromatic activity: R - G and B - (R
+ G) (or the reverse). Taking cone cells of the macaques to be comparable to
those of humans, they analyzed the results in terms of Smith and Pokorny’s
fundamental cell responses and identified three cardinal planes in a space
taken to be euclidean. The constant luminance plane has a constant R + G
(cone response) axis and perpendicular to it a constant B axis. Perpendicular
to the resulting plane is the luminance axis (Fig. 6-21). Colors are defined by
their azimuth F and elevation Q in the space. Derrington et al. determined the
 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS OF COLOR VISION 257

Fig. 6-19 Colors of the 1929 Munsell Book of Color as viewed under CIE illuminant C in
Cohen’s RLV fundamental color space (Burns et al., 1990). The spectral trace is also shown
as are the vectors of the equal energy illuminant and illuminant C.

Fig. 6-20 Spectral functions representing the red, violet and luminosity axes of the RLV version
of Cohen’s fundamental color space (Burns et al., 1990).
258 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-21 Schematic view of the DKL color space based on cone sensitivity data. A color (black
dot) is represented by its azimuth, F, and its elevation, Q (Derrington et al., 1984.)

location of the chromatic axes in the CIE chromaticity diagram (Fig. 6-22).
This space has direct neurophysiological support from measured cell activity
in the LGN. The axes are neither in agreement with average unique hues nor
with those of the CIE-based opponent color chromatic diagram.

Color vision models that imply a color space have also been developed in con-
nection with color appearance models. An inclusion of two of these (Hunt and
Nayatani) in the Mahy et al. (1994) evaluation of uniform color spaces indi-
cates that they represent the Munsell value 5 plane and the OSA-UCS systems
with significant deviations. Such systems, as mentioned earlier, are considered
outside the scope of this text.

De Valois and De Valois

In 1993 R. L. and K. K. De Valois presented a paper titled A multi-stage color
model. They had been involved in some of the earliest experiments identify-
ing opponent color type cells in the LGN of macaques in the 1960s. They were
cognizant of the discrepancy between the output of opponent cells in the LGN
and psychological scaling of color space touched on above. The multi-stage
model was developed to account for such discrepancies. The De Valois model
consists of four stages. In the first stage, three cone types provide responses to
light striking them, and the authors used the Smith-Pokorny functions to
 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS OF COLOR VISION 259

Fig. 6-22 Axes of constant R&G and constant B of the DKL space in the CIE chromaticity
diagram. The squares represent the chromaticities of the phosphors used in the experiment.

describe cone sensitivity. They assumed proportions in the retina of L, M, and

S cones in a ratio of 10 : 5 : 1. In the second stage, cone opponency signals are
generated. Two kinds of surround are considered here: an indiscriminate
surround based on the output of horizontal cells in the retina and the other
a cone type specific surround. The response functions of three cone opponent
cell types, illustrated in Fig. 6-23, are derived as follows:

L0 = L - (L + M + S),

M0 = M - (L + M + S),

S0 = S - (L + M + S). (6-55)

These cells also have mirror image copies. The cells are considered to carry
both luminance and color information at different spatial frequencies.
In the third stage, in the parvocellular pathway, the authors’ proposal posits
combinations of signals in a way that separates luminance and color informa-
tion. Accordingly, for example, L0 - M0 sums color and cancels luminance,
while L0 + M0 sums luminance and cancels color. The response functions of the
third stage are illustrated in Fig. 6-24. They resemble somewhat the Hurvich-
Jameson opponent color functions but are not balanced. They are calculated,
in the indiscriminate version, according to
260 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-23 Three cone opponency signals of the second stage of the De Valois and De Valois
color model (indiscriminate surround). From De Valois and De Valois (1996).

Fig. 6-24 Third-stage chromatic functions of the De Valois and De Valois color model
(indiscriminate surround). From De Valois and De Valois (1993).

RG = 18L - 23M + 5S,

YB = -26L + 19 M + 7S. (6-56)

For the fourth stage, the De Valois proposal has complex color selective cells
resulting from the summing of the responses of simpler cells earlier in the
 COLOR SPACE FORMULAS AND COMPREHENSIVE MODELS OF COLOR VISION 261

pathway. The authors believe that simple opponency such as posited by Hering
and Hurvich and Jameson ends at the LGN level and that already in area V1
of the brain’s visual system there are cells that fire to stimulation from spe-
cific spectral regions and not to those of others. “The chromatic opponency at
this stage is between, not within, individual cells.” At this stage two cell types
each, such as “red” and “blue” or “red” and “yellow,” can fire at the same time
to the same stimulus, resulting in mixed hues. In the form presented here, the
De Valois model does not result in a close representation of the Munsell
system.
The De Valois model is of particular interest because it considers a wealth
of neurophysiological data accumulated over the last fifty years. It represents
informed assumptions in regard to how the visual system operates (up until
area V4) and is in this respect different from Guth’s early models that
optimized the Müller structure to psychophysical data, as the authors point
out in a commentary in 1996.
In 1997 De Valois et al. reported on a hue-naming experiment whose results
they explained in terms of a modified version of their model where in the third
stage they subtracted modified amounts of second stage S0 from L0 and M0
(roughly comparable to the change from the a, b diagram to the a, b diagram;
see Section 5.7) They also found the red and the green systems to likely receive
non-symmetrical inputs from cones or LGN cells. The De Valois proposal is a
recent example of a color vision model that assumes a relatively simple rela-
tionship between proposed outputs of cells in the visual area of the cortex and
color perceptions.

The Opponent Color System: Asymmetrical?

In 1999 E. J. Chichilnisky and B. A. Wandell added data to the thesis that the
opponent color system might be asymmetrical. They conducted an experiment
in which square color stimuli of 2.5° visual angle on a CRT were briefly flashed
against various backgrounds. Observers were asked to identify if the stimuli
were greenish or reddish, yellowish or bluish. One type of their presentation
of the results is in cone contrast space as differences in L, M, and S as opposed
to the corresponding values of the surround. Significant differences were found
between the three observers in the experiment, but all showed classification
boundaries that were bent, usually in a shallow bowl shape.The bowl shape did
not in all cases pass through the origin; that is, “opponent classifications are not
based exclusively on the difference between test stimulus and background
light.” An example of the results of the redness-greenness classification bound-
ary against a greenish appearing surround is illustrated in Fig. 6-25. The shape
of the bowl usually has various bulges, indicating not just global but also local
nonlinearities of the boundaries between unique hues. The results were fitted
with what the authors call an increment-decrement opponency model, indi-
cating that increments and decrements in cone absorptions from those of the
surround were treated differently. (Nonsymmetrical effects of increments and
decrements are known from other experiments, as mentioned in Chapter 5.) A
262 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-25 Redness/greenness classification boundary against a greenish surround in the L,

M, S cone contrast space. From Chichilnisky and Wandell (1999).

Fig. 6-26 Schematic view of the color vision model of Chichilniski and Wandell to account
for their hue classification results. Output from cones L, M, S is split into increment/decrement
portions that are separately modulated to form the intermediate cone signals L*, M*, S*. These
are linearly combined to form incremental and decremental pre-opponent signals. They, in turn,
are again linearly combined to form the opponent signals RG, BY, and WB. From Chichilnisky
and Wandell (1999).

schematic view of the model is found in Fig. 6-26. Linear cone signals L, M, S
are are treated separately as increments or decrements with respect to a neutral
point. These are scaled through gain control depending only on the surround.
The resulting intermediate cone signals L*, M*, and S* are combined linearly
and then separated into pairs of increment and decrement pre-opponent
 IS THE OPPONENT COLOR SYSTEM “SOFT-WIRED”? 263

signals. These are combined linearly to create the final opponent color signals,
whose signs determine the opponent color invoked. Concerning the neural
location of opponent color null surfaces the authors comment: “The non-
planarity of opponent color classification boundaries raise the possibility that
either individual parvocellular LGN neurons are not the neural substrate of
perceptual opponency, or that significant nonlinearities in neural responses
were overlooked.” Increments and decrements have also been found to be
treated differently in spatial patterns (Bäuml, 2002).
The relevance of these findings for a model of a global object color space
viewed under conditions much different from those of the described experi-
ment remains to be determined. There is the likelihood that all models using
simple subtractions of cone absorptions or tristimulus values are much sim-
plified approximations of the real mechanisms of color vision. The issue of the
strategy pursued by evolution in the development of the primate visual system
assumes key importance as well as if perception ultimately can be modeled
from implicit cone responses.

6.18 IS THE OPPONENT COLOR SYSTEM “SOFT-WIRED”?

It is an accepted supposition of color psychology that reddish-greenish colors

and yellowish-bluish colors are not possible. However, this has been ques-
tioned since 1983 when H. D. Crane and T. P. Piantanida published a paper
“On seeing reddish green and yellowish blue.” More recently Billock and co-
workers (2001) reported on experiments where observers viewed fields of
opposing colors at personal equiluminance for each observer under image
stabilization (the image impacted the same retinal region at all times). Four
out of seven observers experienced under these conditions homogeneous
mixtures of red and green or yellow and blue (some only after several trials).
The authors proposed a “winner take all” multi-stage model of vision in
which opponency is not hardwired into certain types of cells but is the result
of the combined output of many cells. As a result one or the other of the
unique hues (the winner) takes over. This soft assignment of opponent hues
can be defeated in certain observation conditions resulting in mixed opponent
hue perceptions.

Models of human color vision must be able to explain vast amounts of

different experimental data. All models are empirical because the translation
of chemical signals in area V4 to perceptual experiences is unknown. There
is a likelihood that the strategies generally pursued so far will not succeed in
explaining all visual data because the visual system may pursue empirical
strategies that do not directly follow from the identified mechanisms in the
retina, the LGN and the early visual areas but are the statistical results of past
experiences, possibly elaborated beyond the classical visual areas.
264 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

6.19 SPECTRAL SPACES

As briefly mentioned in Chapter 2, regular spaces of reflectance or spectral

power data can also be formed by content analysis and dimensionality reduc-
tion different from that implied in the color matching functions. When com-
paring the reflectance functions of Munsell chips of constant chroma and value,
regularly spaced around the hue circuit, they are found to change in a regular
fashion. In Fig. 6-27a through d the reflectance functions of twenty (every
second) Munsell chips at constant value and chroma representing a hue circle
are shown. Given the possibility of many metamers for each chip, this fairly
regular change is in part due to the limited number of pigments with their own
specific spectral function shape and in part is a reflection of the (relativized)
hue change between chips. Lenz and co-workers (1996) have calculated princi-
pal components of the reflectance functions of 1269 Munsell chips (including
those of Fig. 6-27) by forming a correlation matrix and determining its eigen-
vectors. The first three eigenvectors are illustrated in Fig. 6-28. It is evident that
the first vector, accounting for over 80% of the correlation between the func-
tions in the total set, has all positive values and represents in a general way the
average height of the reflectance function, roughly indicative of lightness. The
second and third eigenvectors have both positive and negative values and have
a crude resemblance to opponent color curves (compare to Fig. 5-13).The func-
tions are orthogonal to each other. When the twenty reflectances of Fig. 6-27
are plotted, they do not form a circle on a flat plane in the resulting space but
an irregular three-dimensional contour as shown in Fig. 6-29a and b. The drop
lines in Fig. 6-29a provide an indication of the variation in terms of the first
eigenvector. It is evident that this space orders Munsell colors in an ordinal
manner only in that it places the chips in a sequence that is in agreement with
the perceptual sequence. Colors varying in chroma or value (not shown) are
also placed relative to each other in proper ordinal order. However, the space
is far from uniform in terms of the Munsell system. Ordinal order can be also
obtained with many arbitrary continuous functions and does not in itself indi-
cate a useful color space. Other dimensionality reduction methods do not place
such color series always in ordinal order (Ramanat et al., 2003). Eigenvector
spaces depend on the color stimuli and the sample distribution. Eigenvectors
determined from different collections of color samples will be different. In
addition metameric color samples do not plot in the same location in eigen-
vector spaces. For these reasons it is not appropriate to call such spaces color
spaces.

6.20 PERFORMANCE COMPARISON OF VARIOUS FORMULAS

Over the last hundred years color discrimination data have been established
at four different levels of difference:
 PERFORMANCE COMPARISON OF VARIOUS FORMULAS
265
Fig. 6-27a–d Sequential plot of spectral reflectance functions of every second Munsell chip around the hue circle at value 6 and
chroma 8, indicating the comparatively regular changes in reflectance. Note the common crossovers between 550 and 600 nm.
266 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

Fig. 6-28 First three eigenvectors representative of the first three principal components, fitted
to a database of reflectance functions of 1269 Munsell chips. Solid line: eigenvector 1; dashed
line: eigenvector 2; dotted line: eigenvector 3. From Lenz et al. (1996).

1. Subthreshold. These are color-matching error data, as determined by

MacAdam and others.
2. Threshold or just noticeable difference data. Examples are luminance
difference thresholds, wavelength discrimination data, purity threshold
data, and the Wright, Richter, and Witt threshold data.
3. Suprathreshold small differences. These include the RIT-DuPont, Witt
data, and others.
4. Large color difference data. Examples are the Munsell and OSA-UCS
data, and the Guan and Luo data.

Ideally all these data sets would be describable with high accuracy using a
single model/formula. It is obvious from past discussions that this is impos-
sible. When comparing various models of color spacing with typical data sets,
significant variation in results is obtained. In a few cases formulas have been
optimized against a particular data set, and not surprisingly, these formulas
tend to perform better than others against these data. Typical examples are
the FMC formulas optimized against the MacAdam data, the OSA-UCS
formula optimized against the corresponding large difference data, and the
BFD formula optimized mainly against the Luo and Rigg modified data set.
 PERFORMANCE COMPARISON OF VARIOUS FORMULAS 267

(a)

(b)
Fig. 6-29a and b Plots of the twenty Munsell colors of Fig. 6-27 in the space formed by the
three eigenvectors of Fig. 6-28 and in the diagram formed by the second and third eigenvec-
tors. Hue designations of four colors are shown for identification. It is apparent that the space
orders the Munsell colors regularly in an ordinal sense, but it is far from uniform.
268 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

TABLE 6-1 Prediction accuracy of three color difference formulas for various sets of
color difference data
Data set Coefficient of Variation, %
CIELAB CMC BFD
MacAdam, 24 ellipses 34.5 32.2 27.7
Wyszecki-Fielder/GW, 28 ellipses 26.9 20.2 18.8
Witt threshold, 5 ellipses 39.3 39.1 42.6
Luo-Rigg, 131 ellipses 25.1 16.2 13.6
Cheung-Rigg, 5 ellipses 31.6 39.1 42.6
Munsell value, 232 differences 7.3 29.0 21.5
Munsell hue, 365 differences 64.6 50.7 38.4
Munsell chroma, 356 differences 20.5 35.5 41.8
Source: Adapted from Melgosa et al. (1992).

In 1992 Melgosa and co-workers compared a number of data sets using the
CIELAB, CIELUV, CMC, and BFD formulas. From their extensive data
only a few examples are shown in Table 4-1. Results, in form of coefficients of
variation for the difference between visual and calculated differences, are
bewildering. Some conclusions can be drawn from the table:

1. There are systematic differences between the MacAdam data, on the one
hand, and the Wyszecki-Fielder/observer GW data, on the other.
2. Witt threshold data are poorly predicted by any of the three formulas.
3. CMC and BFD are significant improvements over CIELAB for the
Luo-Rigg data but not for the Cheung-Rigg data.
4. CIELAB is a good predictor of Munsell value but a poor predictor of
Munsell hue differences, while CMC and BFD are poorer in predicting
value differences but better (if still poor) in predicting hue differences.
CIELAB is a better predictor of Munsell chroma differences than the
other two formulas (in agreement with findings in Chapter 8 of absence
of chromatic crispening in color differences of the size of Munsell system
differences).

Another view of the progress of color difference formulas in predicting

average visually judged small color differences and of color space formulas in
modeling color solid data is offered in Table 6-2 and in Figs. 6-30 and 6-31.
Typical data sets have been used for this purpose. The small color difference
data set selected is the RIT-DuPont set with 156 difference pairs in 18 color
centers. When the data were established, considerable care was taken to
control significant aspects of the experiment. The nature of the experiment
resulted in all color differences in this data set having a visual difference of
one unit and therefore should have calculated differences of one unit. An
appropriate statistic of the performance of color difference formulas for this
data set is, again, the coefficient of variation. In Table 6-2 coefficients of
 PERFORMANCE COMPARISON OF VARIOUS FORMULAS 269

TABLE 6-2 Comparison of performance of some color-

difference formulas against the RIT-DuPont data set
Formula Coefficient of Variation, %
CIELAB 35
CIELUV 39
SVF 38
BFD 28
CIE94 21
CIEDE2000 20
Kuehni optimization of CIE94 18

Fig. 6-30 Munsell Renotation colors at value 5 plotted in the chromatic diagram of Richter’s
LABHNU formula. From Mahy et al. (1994).

Fig. 6-31 Munsell Renotation colors at value 5 plotted in the chromatic diagram of Hunt’s 1991
color appearance model. From Mahy et al. (1994).
270 HISTORICAL DEVELOPMENT OF COLOR SPACE AND COLOR DIFFERENCE FORMULAS

variation for color differences calculated by various formulas for the set are
listed.
The variation in calculated differences for the data set selected is now
approximately half of that for the 1976 CIE formulas. The largest improve-
ment was achieved, as indicated earlier, by applying variable weights to the
implicit lightness, chroma, and hue differences to deal with chromatic crispen-
ing and the non-euclidean nature of uniform color space (in the absence of
euclidization procedure).
The Munsell system was selected for the purpose of comparison of the
performance of color space formulas. Constant value, respectively metric light-
ness, planes are illustrated. To be in perfect agreement with the perceptual
results, a color space formula should represent an equal value plane of the
Munsell system as a series of concentric circles with equal radial spacing and
radial lines that are equidistant from each other in terms of hue angle. In the
evaluation of Mahy et al. (1994) the formulas (then known) most accurate in
modeling the Munsell system were the LABHNU formula of Richter (a some-
what different version from the one documented above, but found in the same
reference), SVF, and the CIELAB formula (for OSA-UCS it is the OSA
formula and SVF). A significant further improvement will be shown in
Chapter 7 (compare Fig. 7-6). Formulas such as CIE94 cannot be used as color
space formulas because of the variable weights on the difference components,
unless the weights are integrated such as in the Rohner and Rich, Völz,
Thomsen, or DIN99 proposals (see above). It is evident, however, that they
would perform poorer than the formulas above against the Munsell and OSA-
UCS systems because most of the integrated weight of SC relates to chromatic
crispening, absent in the Munsell system and OSA-UCS.
We have seen a rich tapestry of efforts to develop psychophysical models
of color space and color differences applicable to differences of various size.
But as discussed in Chapter 4, we do not have extensive reliable, replicated
data of which we can be confident that they accurately describe color spacing
for a given situation of illumination, surround, size of differences, and the truly
average observer. Various data sets developed at different times vary consid-
erably for generally unknown reasons and are described optimally by differ-
ent formulas. It is important to develop data sets that can be reproduced in
different locations by observer groups of comparable and known relation to
a truly average standard observer (that may still need to be developed). If this
proves to be impossible, it may be due to irreproducible cognitive input into
the judgments. Truly reliable data sets and further knowledge of the human
color vision system may make it possible to develop a psychophysical basis
model, perhaps non-euclidean and certainly nonlinear, that can, with various
parameters, accurately describe the tiling of color space uniform under closely
defined observation conditions. There is a considerable way to get there and
it is evident that CIEDE2000 is a mere milestone on that path.
 Chapter 7
Major Color Order
Systems and Their
Psychophysical Structure

In this chapter only the Munsell, OSA-UCS, and Swedish NCS systems will
be discussed. The former two are the most important attempts to create psy-
chologically uniform systems, the latter uses the presumed natural approach
of Hering (see Chapter 2) to create a color order system, having a regular
structure, but not one uniform in terms of size of perceived differences. There
are several other newer color order systems extant, but they neither make
claims for uniformity nor for regularity according to new, significant psycho-
logical attributes.
The issue of viewing conditions for these systems has been attended to in
different ways.As discussed previously, the Munsell system is illustrated as if the
chips at each value level would be viewed against an achromatic surround of the
same value. Chips of two adjacent value levels are illustrated as if viewed against
an achromatic surround of intermediate value. The actual atlas displays the
chips on white paper (historically of various degrees of whiteness), thus result-
ing in distortions of the value scale, particularly at lower values.The OSA-UCS
system is defined for an achromatic surround of luminous reflectance Y = 30
(L = 0).The atlas samples are in transparent jackets. NCS, finally, has been estab-
lished against an immediate achromatic surround of Y = 78 in a light booth
painted with an achromatic gray of Y = 54. The atlas displays samples on white
paper. Both latter systems only result in the intended color experiences when
viewed against the appropriate surrounds. Munsell and NCS are defined for

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

271
272 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

CIE daylight C and the 2° standard observer. OSA-UCS is defined for daylight
D65 and the 10° standard observer.

7.1 THE MUNSELL COLOR SYSTEM

Development of the System

Albert H. Munsell was educated as an artist and art instructor. His initial inter-
est in color order (beginning in the 1880s) resulted from the need for an edu-
cational tool for instruction in color order for school children and art students
as well as a tool for objectively expressing harmonious color relations.1 His
initial concept employed a sphere.The inspiration for the sphere form (Munsell
became aware of Runge’s book only in 1899) came from a child’s ball with
colored segments and from plotting the colors of one of his paintings in form
of a double spiral, suggesting a sphere (Munsell, 1918). Munsell, having con-
vinced himself that Hering’s approach could not be correct, decided to base
his hue circle on five primary colors: red, yellow, green, blue, and purple. The
main motivation was to be able to express his system in a decimal framework.
Munsell, not burdened with detailed knowledge about earlier attempts
at systematic color order, unhesitatingly decided to make lightness (“value” in
his artistically influenced term) a key attribute. His early sphere models, for
which he had received a patent in 1900, placed hues of equal value on the
equator, that is, in addition to middle green and red, darkened yellow, and light-
ened blue and purple. Munsell invented the term “chroma” to designate the
radial dimension from the neutral gray center to the equatorial colors. After
plotting “intensity” against “luminosity” values of painted pigments, measured
by Abney, Munsell realized in the year 1900 that on the basis of attributes hue,
value, and chroma, a uniform color solid as represented by available pigments
could not fit into a sphere but would form an irregular “spheroid.” As a result
he abandoned the idea of his color solid fitting into an ideal geometrical solid.
Munsell named the irregularly shaped solid a “color tree.” He maintained the
color sphere as an educational tool until the end of his life, and it remained a
part of the descriptive literature of the system until the Second World War (Fig.
7-1). Munsell began preparing an equal value chart of painted paper chips in
1901 and proceeded with charts at other value levels. In 1902 he sketched a
model based on constant hue planes and began to assemble corresponding hue
charts (Fig. 7-2). During the same years Munsell decided that the system should
be visually uniform, that is, steps along the three attributes should, within an
attribute, be of equal perceptual size. Having differences of equal perceptual
magnitude within all three attributes was contemplated by Munsell but
somehow never implemented. In 1904, when preparing a publication describ-
ing the system and as part of a patent application, Munsell also had settled on
his hue, value, and chroma terminology and color identification scheme. Over
a span of five years Munsell had developed all key concepts of the system as
they stand today. In 1905 he published his conceptual description of the system
 THE MUNSELL COLOR SYSTEM 273

Fig. 7-1 Munsell’s irregular “color tree” enveloping the original sphere. From Derefeldt (1991).

Fig. 7-2 Trace of Munsell’s sketch in his color Diary of an irregular “color tree” with constant
hue leafs, dated March 20, 1902.
274 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

under the title A color notation (Munsell, 1905). He was granted in 1906 a patent
for the system, and he received copyright protection for the charts. In 1907 the
first version of the Atlas of the Munsell color system was published, containing
eight charts with painted samples (Munsell, 1907). A second, enlarged version
of the Atlas appeared in 1915 (Munsell, 1915). It consists of 15 charts, five con-
stant hue and seven constant value charts, as well as three charts of general
descriptive nature.2
In the year of Munsell’s death, 1918, the Munsell Color Company was
formed and continued operating under his son. Scientific support was obtained
from the National Bureau of Standards, and the company soon moved from
Boston to Baltimore to be closer to the Bureau. One of the young researchers
at the Bureau was D. B. Judd. Research on uniform spacing of hue, value, and
chroma was continued, and in 1929 the company issued the first version of
the Munsell Book of Color. The Munsell Book of Color is internationally today
perhaps the most widely known color order/appearance system and is com-
mercially available in a matte and a glossy chip edition, with a supplementary
collection of near-neutral color chips. There is also a textile fabric edition.
The history of the Munsell system has been described by Nickerson (1940,
1969, 1976), Berns and Billmeyer (1985), and Kuehni (2002). A modern
description of the system is that by Long and Luke (2001).

The 1915 Atlas

The atlas consists of a text page with four sketches illustrating the color sphere,
the color tree, and schematically the position of the vertical and horizontal
section charts. The first chart (H) consists of chips of the five primary and the
five intermediate hues at values 2 to 8. Chart V consists of the ten-step gray
scale and chips of the highest chroma colors of the five primary hues at their
appropriate value level. Chart C illustrates chroma scales of the five primary
hues shown at their value level. Five charts illustrating vertical cross sections
through the solid with always two hues illustrated per chart follow. Finally,
there are seven star-shaped horizontal section charts illustrating ten hues at a
given value level in all chroma steps possible with the pigments in use.
Fifteen of these chips were measured at the National Bureau of Standards
in 1919, and CIE tristimulus values were calculated at a later date. Additional
70 samples were measured in 1926, again with tristimulus values calculated
later (Gibson and Nickerson, 1940). When plotted, they do not follow a sys-
tematic pattern. Recently measurements of the 58 value 6 color chips of a copy
of the 1915 atlas were made by the author. They are illustrated in Fig. 7-3 in
the aŸ, bŸ diagram (fitted to the Munsell Renotations; see equation 7-1 below).
The arrangement is reasonably regular. In a few cases the highest chroma
level is insufficiently spaced from the previous level. The luminous reflectance
values of the samples were found to vary from approximately 36 to 41.
Reflectances of the value (gray) scale samples had been measured in 1916 and
1926. The results of the value scale measurements of the 1915 Atlas are com-
 THE MUNSELL COLOR SYSTEM 275

Fig. 7-3 Location in the aŸ, bŸ chromatic diagram of the fifty colors of chart 50 (value 5) of the
1915 Atlas, 2° observer and equal energy illuminant. Radial lines connect colors of constant
hue, circular lines colors of constant chroma.

pared in Fig. 7-4 against measurements of the value samples of the 1929 Book
of Color (see below). The scales are somewhat uneven and show breaks in
continuity: the 1915 Atlas value scale between value steps 4 and 5, the Book
of Color scale between steps 5 and 6. They are indicative of the lightness
crispening effect. The 1915 Atlas measurements are optimally linearized with
a single function by applying a power of 0.43 to the luminous reflectance
values.

1929 Book of Color

Based on research conducted together with the National Bureau of Standards,
the Munsell Color Company issued in 1929 a revised and extended version of
its catalog of sample chips. To distinguish it from the Atlas, the new version
was called Munsell Book of Color (BOC). It consists of twenty hues in values
typically from 2 to 8 and chromas from 2 to (in a single case) 14, with a total
of 400 samples.
Their reflectances were measured against magnesium oxide in 1935 by J. T.
Glenn and J. T. Killian, with MacAdam’s help, and the resulting colorimetric
data were published in 1940 (Glenn and Killian, 1940). All samples at value 6
have been plotted in Fig. 7-5 in the aŸ, bŸ diagram. In comparing this figure
with Fig. 7-3, we find considerable differences between the two versions, par-
276 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-4 Luminous reflectance Y of the value scales of the 1915 Munsell Atlas (solid line) and
the 1929 Book of Color (dotted line), 2° observer.

Fig. 7-5 Location in the aŸ, bŸ chromatic diagram of the colors of the value 6 plane of the 1929
Munsell Book of Color, 2° observer and equal energy illuminant.
 THE MUNSELL COLOR SYSTEM 277

ticularly in regard to chroma. The luminous reflectances of the value scale of

the Atlas shown in Fig. 7-4 require, for optimal linearization over the whole
range, a power of 0.40 but show evidence of lightness crispening.

The Munsell Renotations

In the mid-1930s the Optical Society of America formed its Subcommittee on
the Spacing of Munsell Colors, chaired by Newhall. In 1940 it released a pre-
liminary report describing the psychophysical methods used and preliminary
smoothing results for chroma scales (Newhall, 1940). The plotted Glenn and
Killian measurements of the 1929 BOC samples and other evidence had
indicated jaggedness of constant chroma contours in the CIE chromaticity
diagram and, to some extent, of the constant hue lines. In support of the
subcommittee’s work, 41 subjects had made estimates of the magnitude of
differences between gray scale chips of the 1929 BOC to be used for a rede-
finition of the value scale. All visual estimates were made against a white, a
midgray, and a black background (see Fig. 5-16). The report lists averages and
ranges of the judgments. A significant and systematic effect of the surround
on value judgments was noted. A change in continuity of the value function
indicative of the lightness crispening effect in both the DR versus R function
(where R is luminous reflectance) determined by just noticeable difference
measurement and R versus value function from consecutive halving of the
scale had also been reported by Munsell et al. (1933), as mentioned in
Chapter 5.
In 1943 the committee released its final report (Newhall et al., 1943). Minor
irregularities in the mean revised experimental spacing results were smoothed
in the CIE chromaticity diagram. The lightness crispening effect in the value
scale against the gray surround was also smoothed away. Trends in hue and
chroma were extrapolated to the object color limits of the x, y, Y psy-
chophysical color space. The resulting Renotations, expressed for the CIE 2°
observer and illuminant C, define aim values for 2746 chromatic and nine
achromatic colors. Approximately 65% of these have been physically realized
as color chips in two volumes of the BOC. The committee used Judd’s defin-
ition of the value function in form of a smooth quintic equation relating value
to luminous reflectance (see equation 5-8). The Munsell Renotation data have
also been published in Wyszecki and Stiles (1982).
Aim colors at every second of the 40 defined hues of the Renotations
(5 and 10) up to chroma 14 at value 6 are plotted in Fig. 7-6 in the aŸ, bŸ
diagram. The changes compared to the 1915 Atlas and the 1929 BOC are
clearly apparent.
A software package (Leonardo 2000) has been released containing
reflectance functions for the Renotation colors achievable with pigments. It
allows the calculation of colorimetric aim points for both standard observers
and any defined illuminant. It has a display facility that reproduces on the
monitor any such generated Munsell aim color (Brill et al., 2001).
278 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-6 Munsell Renotation aim colors for the 2° observer and equal energy illuminant. Every
second hue (5 and 10) at value 6 and chroma 0–12 in the aŸ, bŸ chromatic diagram. Compare
with Figs. 6-30 and 6-31.

The Munsell Re-renotations

During the 1950s and 1960s the Committee on Uniform Color Scales of the
Optical Society of America, under the leadership of Judd, conducted a series of
experiments with the goal of developing an improved uniform color solid and
associated space formula. Judd summarized these experiments in four reports
(Judd, 1955, 1957, 1965, 1967). Judd and Nickerson were interested in the impact
of the results of these experiments on the Munsell system. As discussed below,
the committee pursued a different path, ultimately resulting in the Society’s
Uniform Color Scales. In 1967 Judd and Nickerson published a National Bureau
of Standards report in which they proposed revised Renotations that reflected
the new experimental findings. They specified a total of 2874 chromatic
colors, in a complete revision of the Munsell system (Judd and Nickerson, 1967).
Even though only a small number of corresponding samples were produced
for experimental purposes, it is instructive to review the Re-renotations to gain
insight into the work of the committee. The same aim colors (by name) of the
Renotations used for Fig. 7-6 are illustrated for the Re-renotations in Fig. 7-7.
It is evident that they represent a significant change.
 THE MUNSELL COLOR SYSTEM 279

Fig. 7-7 Munsell Re-renotation aim colors for the 2° observer and equal energy illuminant.
Every second hue (5 and 10) at value 6 and chroma 0–12 in the aŸ, bŸ chromatic diagram.
Compare with the Renotation colors of Fig. 7-6.

Munsell Psychological Space and Psychophysical Representation

As already sketched in Chapter 2, the ideal form of the Munsell system is cylin-
drical with a uniformly spaced hue circle, chroma, and value scale. We have
seen earlier that this form does not account for the superimportance of hue
nor for the irregular distribution of unique hues. It also is not uniform in
hue terms in that the hue difference between two adjacent planes of constant
hue at constant lightness varies as a function of chroma but not in the way the
constant hue angle between the two planes indicates (see below). A concep-
tual view of the Munsell system is shown in Fig. 7-8. From the discussion above
it is evident that this figure is inaccurate in regard to hue difference in terms
of the basic goals of uniformity of difference, at least within an attribute. To
project the reflectance functions into a space in such a way that the cylindri-
cal form of Fig. 7-8 is duplicated has been a long-standing goal. Many attempts
have been made since the definition of the Munsell Renotations, as has been
seen in Chapter 6. The more successful ones are based on a form of opponent
color system. Such systems usually make use of the CIE colorimetric sys-
tem, the best-known example being the CIELAB color space formula.
The CIELAB formula is not a very good representation of the Munsell
280 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-8 Schematic view of the cylindrical organization of the Munsell system, illustrating four
constant hue planes and the organization of samples on these planes. The chromatic plane is
not representative of the perceptual relations in the Munsell system. From Judd and Wyszecki
(1975).

psychological space as illustrated in Fig. 7-9 for the colors of Fig. 7-6. Among
other things, it also has the same weakness in regard to hue difference as the
psychological conceptual system of Fig. 7-8.
Among the assumptions implicit in most psychophysical models are: (1)
colors are defined by two opponent color coordinates a and b and luminous
reflectance and (2) unit color difference contours can be expressed in terms
of distances in a and b. The CIELAB formula makes the additional assump-
tion that the power modulation of the tristimulus values used to calculate a*,
b*, and L* is uniform and its value is cube root. As discussed before, this
assumption was made by Adams in 1942 and is still in use. The b* axis result-
ing from application of the 2° observer is a reasonable representation of
unique yellow and blue, but as we have seen, the a* axis does not coincide
with unique green and unique red. Disregarding this difficulty, we can take the
 THE MUNSELL COLOR SYSTEM 281

Fig. 7-9 Representation of the colors of Fig. 7-6 in the CIELAB a*, b* diagram, 2° observer
and equal energy illuminant.

a* scale to approximately represent greenness-redness and the b* scale

yellowness-blueness. In the four quadrants of the a*, b* diagram the number
and spacing of hue difference steps of equal size varies.
The power modulation implicit in the Renotations can be investigated by
checking it along the axes. Because adaptation negates most of the effect
of different broadband white light illuminants, tristimulus values are, for pur-
poses of simplification, normalized to the equal energy illuminant. By defini-
tion, colors along the chromatic axes at constant luminous reflectance change
in X, respectively Z, only. When plotting these values, we find that the power
modulation required to optimally linearize the progression of tristimulus
values of the corresponding nearest Renotation colors differs by semi axis
(Kuehni, 2000b). In Fig. 7-10a and b the progression of X, respectively Z, values
of Munsell Renotation colors at value 6 along or very near to the opponent
axes is illustrated as a function of chroma. The results indicate discontinuous
functions, separated by the neutral gray. The optimized power functions for
the four segments are as follows (as shown in Table 5-2): a - (green) colors
0.19, a + (red) colors 0.15, b - (blue) colors: 0.06, b + (yellow) colors 0.42. All
are significantly different from a cube root function. The equations used for
282 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-10a Progression of X tristimulus values of Munsell colors nearest to the a opponent
color axis as a function of Munsell chroma, value 6, 2° observer, and equal energy illuminant.

Fig. 7-10b Progression of Z tristimulus values of Munsell colors nearest to the b opponent
color axis as a function of Munsell chroma, value 6, 2° observer, and equal energy illuminant.
 THE MUNSELL COLOR SYSTEM 283

calculating the chromatic coordinates of the colors of Figs. 7-3, 7-5, 7-6, and
7-7 are as follows:
0.15 0.85
ÈÊ X ˆ ÊYˆ Ê Y ˆ˘
aŸ + = 3500 Í * - ,
ÎË X 0 ¯ Ë Y0 ¯ Ë Y0 ¯ ˙˚
0.19 0.81
Ÿ
ÈÊ X ˆ ÊYˆ Ê Y ˆ˘
a - = 2000 Í * - ,
ÎË X 0 ¯ Ë Y0 ¯ Ë Y0 ¯ ˙˚
0.06 0.94
ÈÊ Y ˆ Ê Z ˆ ÊYˆ ˘
bŸ + = 1400 Í - * ˙,
ÎË Y0 ¯ Ë Z0 ¯ Ë Y0 ¯ ˚
0.42 0.58
ÈÊ Y ˆ Ê Z ˆ ÊYˆ ˘
bŸ - = 485Í - * ˙, (7-1)
ÎË Y0 ¯ Ë Z0 ¯ Ë Y0 ¯ ˚
where X, Y, Z are the CIE tristimulus values of the Munsell colors for the 2°
observer and illuminant C and X0, Y0, Z0 are the tristimulus values of illumi-
nant C. The resulting aŸ and bŸ values are on average of the same size as those
obtained with the CIELAB formula. Without a further, nonlinear scaling
factor the formula is only applicable to value 6.

From the Cone Shape in Tristimulus Space to the Cylinder in

“Uniform” Psychophysical Space
As was seen in Chapter 5, constant chroma circles at different value levels
approximately form an inverted, slanted cone in X, Y, Z tristimulus space.
When translated into the linear opponent color space, an inverted cone
remains but is aligned with the vertical Y dimension of a, b, Y space. Con-
ceptually application of a constant power to tristimulus values efficiently con-
verts this cone into a cylinder. When plotting Munsell Renotation colors
closest to the b axis of the opponent color space (close to unique yellow and
blue for the average observer) colors of the same chroma at different value
levels are found to fall on a straight line (at higher values) that turns into a
curved line at lower values (Fig. 7-11a). The same applies to colors closest to
the a axis (Fig. 7-11b). However, the pattern is different. Much less of the avail-
able area is filled by the bluish green-to-bluish red colors than by the yellow-
to-blue colors. When extrapolating the initially straight lines they are found
not to intersect in one point. Intersection in one point applies if the chroma
spacing at a given level of value is linearized by the same power as the value
scale. But as we have seen, the chroma scales in the four axis directions from
the central gray colors are optimally linearized by different powers. The
Munsell Renotations have curiously uneven first steps from gray in the four
directions. The validity of these steps could not be confirmed (Kuehni 2000c).
Figure 7-11a and b indicate the constraints on the existence of certain colors
at certain value levels.
Figure 7-12 illustrates the Weber fractions of Z at various value levels. The
surprising result is two nonlinear functions of mixed composition, the upper
284 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-11a Plot of Munsell colors nearest to the a axis at six value levels (from the top 9, 8,
6, 3, 2, 1) in the X, Y diagram, 2° observer and equal energy illuminant. Colors of constant
chroma at different value levels are connected by lines.

consisting primarily of higher value colors and the left half of the lower curve
of lower-value colors. As a result of the different powers required for chroma
linearization, a cylinder is not obtained without additional adjustment. Empir-
ically the application of a factor of 0.133Y0.59 was found to provide the required
correction for the Renotations (Kuehni, 1999).

Curvature of Constant Hue Lines in the CIE Chromaticity Diagram

In most cases Renotation colors of constant hue are found to fall on curved
lines in the CIE chromaticity diagram (Fig. 7-13). This is the result of the so-
called Abney effect (Abney, 1910). When varying amounts of an achromatic
light are added to a monochromatic light the resulting hue is not constant but
changes as a function of the relative contents of chromatic and achromatic
light. When such lights or corresponding object colors are represented in a
uniform chromatic diagram constant hue lines become straight. Mathemati-
cal/geometrical relationship between the two methods of presentation pro-
vides an explanation. In a conceptual example an aŸ, bŸ diagram is illustrated
(Fig. 7-14a) in which the colors along the four axes required different powers
 THE MUNSELL COLOR SYSTEM 285

Fig. 7-11b Plot of Munsell colors nearest to the b axis at six value levels (from the top 9, 8,
6, 3, 2, 1) in the Z, Y diagram, 2° observer and equal energy illuminant. Colors of constant
chroma at different value levels are connected by lines.

for optimization (aŸ+: power 0.75, aŸ-: power 0.5, bŸ+: power 0.25, bŸ-: power
0.333).
Calculating backward to a linear a, b diagram (Fig. 7-14b), we find the axis
colors spaced differently and the intermediate colors connected with lines of
different curvature. Curvature of the lines of intermediate color is also obtained
if the colors along an axis are optimally linearized by a single power. If the same
power is required for both axes, the line at 45° is straight while the one at 135°
has a curvature. When calculating the corresponding chromaticity coordinates
(Fig.7-14c) curvature is also obtained. Geometry requires that in order to have
straight lines after linearization with the appropriate power non-axis lines need
to be curved both in the linear a, b and the x, y diagrams in a way that depends
on the powers involved.Thus in the x, y diagram most lines connecting constant
hue colors of the Renotations are curved.

Unit Difference Contours in the Munsell System

Given the information about the relative magnitude of hue, value, and chroma
differences in the Munsell system by Nickerson and by Bellamy and Newhall
286 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-12 Weber fractions DZ/ Z as a function of tristimulus value Z of Munsell colors nearest
to the b axis at six value levels, 2° observer and equal energy illuminant.

and that colors falling on two neighboring constant hue lines vary in hue
differences between them, we can make estimates of the shape of unit chro-
matic differences in the Munsell system. At the level of differences of the
magnitude of Munsell chroma steps, at chroma 5 three Munsell 100 hue steps
equal two chroma steps. On this basis most unit contours in the psychological
diagram are of a rectangular or oval nature (or perhaps trapezoidal) with
the major axis of the contours aligned with constant hue lines. Based on
Nickerson’s relationship at chroma 5 we can calculate a ratio between major
and minor axis of approximately 2 : 1. Using the ratio of Bellamy and Newhall
at threshold the ratio is approximately 2.8 : 1. This indicates that unit contours
at Munsell sample size differences are somewhat less elongated than at thresh-
old level, that is, smaller hue differences require a relatively smaller stimulus
increment than larger hue differences. Judd showed in 1968 that, as a result
of the unit difference contour, a uniform perceptual color space cannot
be mapped isomorphically into a geometric space. A euclidean model of a
uniform space at the level of Munsell differences is impossible. Judd’s answer
to this question was a “crinkled fan” (Fig. 7-15). While a sector of this fan can
be spread out a complete circular crinkled fan cannot. As a result it is appar-
ent that the Munsell system is but a regular approximation of a uniform color
solid.
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES (OSA-UCS) 287

Fig. 7-13 Portion of the CIE chromaticity diagram with radial lines of constant Munsell hue
and ovals of constant chroma, 2° observer and illuminant C. From Agoston (1987).

7.2 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR

SCALES (OSA-UCS)

Development of the System

We have already encountered some aspects of the development of OSA-UCS
in the discussion of the Munsell Re-renotations. In the 1940s the issue of a
uniform color space had the attention of the American National Research
Council and a list of industries and governmental departments that could
profit from the outcome had been established (Judd, 1955). In 1947 the Opti-
cal Society of America decided to undertake this research and a committee
under the chairmanship of Judd (at the National Bureau of Standards) was
appointed. He chaired it until shortly before his death in 1972 whereupon
MacAdam assumed chairmanship. Membership varied somewhat during the
nearly 30 years of its existence but never consisted of more than fifteen people
(Nickerson, 1977). Based on a paper from 1941 by Balinkin, describing regular
288 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 7-14a Hypothetical colors falling on the axis lines and intermediate angles in an aŸ, bŸ
diagram that by semi axis required different powers for linearization.

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 7-14b Back-calculated linear opponent color diagram of the same colors showing
curvature of lines connecting intermediate angle colors.

tetrahedral tiling of a uniform color space, Foss made already in the first formal
meeting of the committee in 1947 a proposal to tile the committee color space
in this manner. As L. Silberstein had pointed out in 1942, if one makes the size
of color differences in a triangle equal in magnitude, there can only be six
nearest neighbors in the chromatic plane for the system to be euclidean. With
more than six neighbors the space assumes hyperbolic shape, with less than
six Riemannian. Silberstein recommended testing hexagonal arrays of colors
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES (OSA-UCS) 289

Publisher's Note:
Permission to reproduce this image
online was not granted by the
copyright holder. Readers are kindly
asked to refer to the printed version
of this chapter.

Fig. 7-14c Colors from Fig. 7-14a in the CIE chromaticity diagram showing a different curva-
ture of intermediate angle color lines.

Fig. 7-15 Segment of Judd’s “crinkled fan” uniform color difference surface which considers
the unit uniform difference contour of the Munsell system as a ratio between major and minor
axes of 2 : 1. The straight line across the open fan is a secant. It is also shown on the crinkled
version. When the crinkled fan is circular (representative of the total Munsell hue circle), it
cannot be spread out (Judd, 1968).
290 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

to determine if color space is euclidean. As we will see, in its experimental

work the committee decided to follow this recommendation.
At the time there were three schools of thought in the committee:

1. The experimental facts of color difference are established, and no further

work in this respect is required.
2. The experimental facts are very complex and appear to prove that a
uniform euclidean color space is not possible. The best goal of the
committee is to develop a simple color difference formula.
3. Uniformity, or its lack, is best exemplified by a set of color chips.

As it progressed the committee developed new experimental data, found that

a uniform euclidean space is not possible, and developed what it considered
to be the best approximation of a uniform euclidean system possible, exem-
plified by a set of chips and a formula.3
By the end of 1954, 38 color studies related to uniform color scaling had
been reviewed (Judd, 1955). In 1952 a decision was made to assemble a
500-chip uniform three-dimensional sample set in a regular rhombohedral
arrangement based on the Munsell Renotations, adjusted for its discrepancies
with the results of the MacAdam (1942) and Brown-MacAdam (1949) color-
matching error data. In preparation Nickerson, Judd, and Nimeroff in 1955–56
developed a constant saturation locus at value 6 (see Fig. 5-22). Newhall deter-
mined hue spacing on this locus in 1957–58, and Howett determined chro-
maticness spacing for near grays at values 3, 6, and 8 as well as chromatic-
ness spacing of the Munsell value 6 plane.
Following Silberstein’s recommendation, a color scaling experiment involv-
ing 43 color samples at near-constant luminous reflectance in a triangular grid
pattern was then undertaken. Two auxiliary experiments were also performed:
(1) determination of the relative size of chroma steps at value 6 to value steps
at the same value level; (2) determination of the value of the gray having the
same perceived lightness as the 43 samples with nominal value 6 (test of the
Helmholtz-Kohlrausch effect). The 43 samples were arranged to form con-
tiguous triangles in the chromatic plane, none involving the achromatic gray
(Fig. 7-16). They are fundamental for the subsequent development of the
Uniform Color Scales. From this arrangement 107 (mixed hue and chroma)
differences between nearby colors resulted. They were evaluated by a total
of 76 observers (different portions of the total set by different numbers of
observers and under presumably somewhat different viewing conditions) in a
two-category forced choice experimental paradigm where the observers deter-
mined if the difference between given samples A and B was larger or smaller
than the difference between samples B and C.4 From the results psycho-
metric scale values were calculated at the NBS. The scale values were evalu-
ated against four color difference formulas: CIE U*V*W*, Glasser-Reilly
cube root, Munsell Renotation (modified Nickerson Index of Fading), and
MacAdam-Friele 1965 (see Chapter 6). The highest correlation coefficient
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES (OSA-UCS) 291

Fig. 7-16 Image of the 43 hexagonal enamel color plates used by the OSA-UCS committee
to establish the fundamental perceptual data for the system. Note that there is no achromatic
color among them. (See color plate.)

between visual and calculated color differences was a disappointing 0.45 and
was achieved with the Munsell formula. A series of iterative steps (12 revi-
sions, 33 steps) was subsequently taken to modify that formula as well as the
Munsell Renotation aim data to improve the correlation between formula and
committee data to 0.80. The resulting revised aim data are the Re-renotations
of the NBS report discussed above.
Reilly developed modifications of the cube root formula to improve cor-
relation with the visual data, and MacAdam proceeded from the MacAdam-
Friele space to a nonlinear transformation space based on geodesics. Reilly
found that correlation was significantly improved by applying a superhue
weight of 2.3. Key results obtained from analysis of the visual data and com-
parison with colorimetric data were as follows:

1. Additivity fails by as much as 30%: ratio judgments of the differences

between neutral gray and constant chroma chromatic colors in opposite
direction found them to be larger by up to 30% than the direct differ-
ence between the two terminal colors.
2. Psychometric scales were found to be nonlinearly related (power 0.37 to
0.80) to chromaticity coordinate differences.
292 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

It had become evident that it was impossible to present the results of the eval-
uation of the 43 samples of the value 6 plane on a flat surface. Thus the results
were in this respect in agreement with the Munsell system and findings of the
MacAdam color-matching error data. The committee decided to proceed
along two lines:

1. Prepare a set of color chips representing the closest a euclidean system

can come to a uniform color solid and thereby to demonstrate “the
reality of the various geometrical difficulties embodied in a set of painted
colors” (Nickerson, 1977).
2. Develop a euclidean mathematical formula that best describes the
experimental findings. At its thirteenth meeting in 1967 the committee
voted to adopt the following statement: “We will affirm to the world that
no regular rhombohedral lattice sampling of color space, with a fixed
background, can exist; we will produce the best approximation to such
a lattice for neutral value 6 background that we can design; and we will
specify some or all of the perceived sizes of the differences in our lattice.”

The interest of design-oriented people in the novel scales revealed by the

crystalline structure was also a consideration for continuing the effort.
The test of the magnitude of the Helmholtz-Kohlrausch effect using a set
of gray tiles to compare against the 43 chromatic tiles gave results in reason-
able agreement with the Sanders and Wyszecki (1957) determination using
light colors (see Chapter 5 for a quantitative comparison of the two sets of
data). The committee developed a formula to describe the results as a func-
tion of the CIE chromaticity coordinates.
To establish additional experimental facts of the relationship between chro-
matic and lightness differences, another four sets of four samples each were
prepared. Each set formed a regular tetrahedron in the tentative “uniform”
color space, one set being yellow at Y = 90, another red at Y = 10, the third
green at Y = 30, and the last blue at Y = 6.6. Differences were compared within
and across sets. A table with the final data lists the observed scale differences
as determined from the psychometric scales, the difference computed with the
committee’s formula and the error (MacAdam, 1974).

Optimized Formula
With the visual results in hand several independent efforts to arrive at an opti-
mized mathematical formula were made. In 1970 two formulas were in con-
tention as the best: a version of MacAdam’s nonlinear transformation formula
(xi, eta formula; see Chapter 6) and Reilly’s modified cube root formula.
MacAdam’s formula was found not to extrapolate well beyond the experi-
mental data. The committee was interested in a formula that extends reason-
ably to the limits of object color perception (the MacAdam limits). Reilly’s
formula was found to do a better job in this respect. Various parameters of
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES (OSA-UCS) 293

Reilly’s formula were now optimized to provide the best correlation obtain-
able with such a formula (see equation 6-54). The committee selected a set of
color fundamentals R, G, B to fit the visual data that imply as yet unknown
neurophysiological processes to achieve the fit, illustrated in Fig. 7-17a. Figure
7-17b shows the committee’s opponent color functions j and g. In Fig. 7-17c
the linear form of these functions is compared to balanced linear opponent
functions a and b derived from the 10° observer functions. For ease of com-
parison the conventional functions have been multiplied by a factor of 8. The
results indicate systematic deviations, particularly in higher implied redness
and lower blueness of bluish-reddish colors (short wavelengths).
The lightness scale is a modified Semmelroth scale based on a surround of
Y = 30 and is illustrated in Fig. 7-18. Lightnesses of chromatic colors are further
adjusted by a correction for the Helmholtz-Kohlrausch effect in the form of a
modified Sanders-Wyszecki formula with calculation from CIE chromaticity
coordinates (it should be noted that experimental determination of the HKE
had been made only at one level of luminous reflectance).
The fit of the formula to the visual data involving 43 samples is shown
graphically in Fig. 7-19 (MacAdam, 1974). The gaps represent cases where the
calculated difference exceeds the average perceived differences, and the rec-
tangles the opposite. It is evident that while there is a general impression of
inhomogeneity, there is a degree of systematic deviation. Individual differ-
ences have errors of the calculated difference of up to 70%. For a mean
observed difference of 2.483 units for all 174 differences included by the com-
mittee the root mean square error (RMS) using the committee’s final formula

Fig. 7-17a Spectral R, G, B functions selected by the Committee on Uniform Color Scales for
the formula fitted to the experimental data, 10° observer and illuminant D65.
294 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-17b Spectral opponent color functions j and g as defined by the Committee, 10°
observer and illuminant D65.

Fig. 7-17c Comparison of linear spectral opponent functions of the OSA-UCS system with
balanced linear functions a and b derived from 10° observer data. For ease of comparison
the linear g function has been inverted and the a and b functions multiplied with the factor 8.
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES (OSA-UCS) 295

Fig. 7-18 OSA-UCS lightness L as a function of luminous reflectance Y, with lightness crispen-
ing for surround L = 0.

was calculated as 0.421, or 17%. The correlation coefficient between calculated

and observed values is 0.45.
The system was described by MacAdam in 1974. An extensive report of the
work of the committee, envisaged by Judd, was never published. It is surpris-
ing to note that the issue of hue superimportance and its implication for
the accuracy of OSA-UCS is absent from the system description. Judd covered
hue superimportance in considerable detail in his 1968 paper on ideal color
space but without reference to OSA-UCS. In his book Color measurement
of 1981 MacAdam made the following statement: “. . . the committee forced
the data into a euclidean form, despite clear indications that the judgment
data required noneuclidean representation.” The issue is absent in Wyszecki
and Stiles’s, Derefeldt’s, or Berns’s recent description of OSA-UCS (Berns,
2000).

OSA-UCS Atlas
With the formula established, the committee proceeded to sample the implied
space according to the rules of regular rhombohedral sampling. Rather than
use triangular sampling of the constant lightness plane, it was decided to rotate
the unit cubo octahedron to obtain a square grid in the constant lightness
296 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-19 Plot of 102 chromatic differences between 43 color samples in the chromatic
diagram of the Committee on Uniform Color Scales. Numbers identify the samples. Lines with
arrows and bars indicate the size of the average visual judgments. The colorimetric difference
is too small in case of arrows, too large in case of bars. From MacAdam (1974).

plane. The twelve distances from the central point M to any of the points on
the surface of the cubo octahedron are the same (Fig. 7-20; see also Figs. 2-49
and 2-50). By implication, the distances between the four points each (I, J, K,
L and A, B, C, D) are also the same. The basic cubo octahedron of the system
is illustrated in Fig. 7-20 and the lattice, doubly expanded in all three dimen-
sions, is illustrated in Fig. 7-21. Preceding and succeeding constant lightness
planes are offset by one j, g unit. An illustrated description of the space lattice
used was offered by Foss in 1978 (see Fig. 7-21).
A three-dimensional model of the system was manufactured by MacAdam.
He mounted spheres, dipped in the formulated paints, in cubo octahedral con-
figuration as shown in Fig. 2-51. This figure, as well as Fig. 7-20, illustrates the
existence of seven cleavage planes that allow bisection of the color solid in
various directions. The cleavage planes are formed by the following points in
the figure:

Plane 1 E, F, G, H (constant lightness plane)

Plane 2 B, F, I, K, H, D
Plane 3 A, F, I, L, H, C
Plane 4 I, J, G, D, C, E
Plane 5 K, L, G, B, A, E
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES (OSA-UCS) 297

Fig. 7-20 Cubo octahedral arrangement of twelve colors equally distant from central color M.
Colors I, J, K, and L represent the square forming the square grid pattern of the OSA-UCS
system at constant lightness. Compare with Figs. 2-49 and 2-50.

Fig. 7-21 Basic cubo octahedral unit lattice doubly expanded in all three dimensions. From
Foss (1978).
298 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Plane 6 I, A, D, L
Plane 7 J, B, C, K

In 1974 MacAdam stated: “The regular-rhombohedral system of color

sampling adopted by the Optical Society Committee on Uniform Color Scales
provides the maximum possible number and variety of uniform color scales
and exhibits the maximum possible variety of relationships among colors. The
committee hopes that artists and designers will find it useful in devising new
and beautiful arrangements of colors.”
An identification system based on three numbers was devised to designate
each color. The letter j (for the French jaune, yellow) roughly indicates
yellowness-blueness, the letter g approximately greenness-redness, and the
letter L lightness. For the samples of the atlas j ranges from -6 (blue) to +12
(yellow), g from -10 (red) to +6 (green), L from -7 (dark) to +5 (light). Simple
mathematical relationships connect the colors of each cleavage plane. In
studying systematic color scales, the committee noticed the paucity of near
neutral samples and decided to add a series of such samples at half steps,
centered on L = 0 (the pastel set). The 134 samples of that set range from
L = -1.5 to L = 1.5.
The Optical Society published in 1977 an atlas containing a total of 558
glossy color chips (424 of the regular and 134 of the pastel set), representing
the regular rhombohedral system. Colorimetric specifications of the aim colors
in terms of CIE chromaticity coordinates and luminous reflectance for the 10°
observer and illuminant D65 have been provided by MacAdam (see also
Wyszecki and Stiles, 1982).
A detailed description of the regular rhombohedral structure of the OSA-
UCS system is found in Agoston (1987). A standard practice for using the
system has been issued as ASTM procedure E1360. A software package, Color
Cleaver®, is available that displays approximate OSA-UCS scales on a color
monitor (Luke, 2001).

A Revised OSA-UCS
In 1990 MacAdam published a revision of OSA-UCS based on 2000 judgments
by two young color normal observers. A major advantage claimed by
MacAdam for the new observations is that for the first time samples of all
twelve lightness planes were evaluated by the same observers. As a result of
regression calculations he proposed redefined R and B values as well as rede-
fined j and g values:

R = 0.9285 X + 0.3251Y - 0.1915Z ,

B = -0.2032 X + 0.60Y + 0.5523Z ,
g = C (12.7R1 3 + 19G1 3 - 6.3B1 3 ),
j = C (-1.3R1 3 + 17G1 3 - 15.7 B1 3 ). (7-2)
 OPTICAL SOCIETY OF AMERICA UNIFORM COLOR SCALES (OSA-UCS) 299

These changed values amount to a significant revision. MacAdam also

calculated revised aim colors for a regular rhombohedral system based on the
revised coordinates.5 A limited number of corresponding samples were pre-
pared in Denmark by Ransing. However, in a handwritten letter dated March
22, 1995, MacAdam stated: “Incidentally, I have found some errors in that
(1990) JOSA paper, and withdraw it entirely.” (MacAdam, 1995) An extensive
visual confirmation of the OSA-UCS scales is missing to date (but see Indow,
2001).
Comparisons of the Munsell Renotations with OSA-UCS have been made
by Wyszecki (1954), Nickerson (e.g., 1978), and Agoston (1987).

How Uniform Is OSA-UCS?

Disregarding hue superimportance in the formula resulted in reduced corre-
lation and, presumably, perceptual nonuniformity of the differences between
the samples. In the late 1980s an Inter-Society Color Council Project
Committee (44) performed scaling near the achromatic point of the OSA-
UCS scales. They found the first steps from gray in the four axis directions
of the even numbered lightness levels to be too large compared to more chro-
matic steps, perhaps indicative of a small chromatic crispening effect. The
results of the committee have not been published, and it has been dissolved.
Recently Indow has begun to psychologically measure OSA-UCS using
comparison against a Munsell value step. Preliminary results (Indow, 2001)
indicate a surprising level of disagreement between the committee results and
those of Indow. The disagreement appears to be random in nature. Visual
differences along constant j and g lines in the L = 0 constant lightness plane
vary from 0.63 to 1.55, a factor of nearly 2.5. The coefficient of variation of all
of Indow’s judgments against the system is 23%. The first steps from gray were
found to be larger in the red and blue directions as compared to the green and
yellow directions, a result not in agreement with that by Kuehni (2000c).
As already Reilly had found, considerable improvement of fit over that
achieved by the committee’s formula is possible by considering hue superim-
portance. Considerable reduction in the RMS error (when compared to visual
data) is obtained by separately weighting by quadrant the hue and chroma
differences. They are calculated for differences between sample pairs of
applying CIELAB methodology to the OSA-UCS total differences. Chroma
and hue differences were calculated assuming a euclidean space. The opti-
mization was limited to the 104 differences involving the original 43 tiles.
Optimal results by the method described were obtained when multiplying in
quadrants 1–3 chroma differences with the factor 0.6 and hue differences
with the factor 1.3. In quadrant 4 the factors 1.3 for chroma differences and
1.8 for hue differences proved optimal. The higher values but smaller ratio for
quadrant 4 (greenish-bluish colors) indicate that here the visual differences
between sample pairs were seen as relatively larger than in the other three
quadrants, for reasons that are not obvious. The factors in quadrants 1–3 indi-
300 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

cate a superhue weight of 2.2 (close to the one obtained by Reilly, mentioned
earlier). The resulting average total color difference was nearly identical to the
committee’s value for these data. For the reduced data set (104 differences)
the committee’s RMS error is 20.2%, with a result of 12.7% after the described
optimization. The correlation coefficient is improved from a value of 0.45 to
0.81. The results indicate that due to superimportance of hue the unit chro-
matic difference contours in the OSA-UCS basis data in the j, g diagram are
also elongated in a radial direction, in agreement with the Munsell and small
color difference data.
More exactly the ratio of the multiplication factors in quadrants 1–3 is 2.17,
comparable to the ratio of 2.0 for Munsell colors at chroma 5 (2.8 at the JND
level). In quadrant 4 it is 1.38. Colors falling on diagonals at 45 and 135° are
chroma differences only. Colors such as j4, g0 and j0, g4 are essentially hue
differences from j2, g2, a color on the diagonal. If hue superimportance has
been suppressed in the physical system, one would expect the essentially hue
differences to be perceptually larger than the chroma differences. Informal
evaluations confirm this. A formal experiment is required. The situation
implies that in the final system the visual distances along diagonals in a square
anchored at the neutral point are not identical as they should be in a uniform
system.
As a result we find that the apparent paradox of circular unit chromatic
contours in OSA-UCS compared to the elongated unit contours in the Munsell
system and in small color difference data is due to the suppression of hue
superimportance in OSA-UCS. In reality the visual data in all three cases are
in general agreement. The suppression has also resulted in unequal visual
distances along diagonal lines in the j, g diagram. As a result of these nonuni-
formities the major value of OSA-UCS should today perhaps be seen in its
various types of more or less uniform scales and their aesthetics.

OSA-UCS Psychological Space and Psychophysical Representation

The form of the OSA-UCS psychological space is that of a cube. The chro-
matic plane is defined by a regular grid of squares. While lightness is an explicit
attribute, hue and hue difference is not. Chroma is only explicitly defined along
the two major axes and the diagonals. All other steps in the chromatic plane
are varying sums of hue and chroma differences. As mentioned before, unlike
Munsell, the OSA-UCS system is defined against a specified surround of Y =
30 and constant lightness of chromatic samples does not imply constant lumi-
nous reflectance because the latter has been adjusted to reflect the Helmholtz-
Kohlrausch effect. The locations of unique hues in OSA-UCS as shown in
Table 4-1 indicate that this system is considerably different from a Hering
system in terms of implied hue spacing.
Plots of X against Y values (Fig. 7-22a and b) of colors along the axis where
Z = 0 and Z against Y where X = 0 reveal a picture similar to that of Fig. 7-
11a and b. However, the spacing is more uniform and continuous throughout
 THE SWEDISH NATURAL COLOR SYSTEM (NCS) 301

Fig. 7-22a OSA-UCS colors falling on the g axis at six levels of lightness L in the X, Y diagram.
Achromatic colors fall on the diagonal, 10° observer and illuminant D65. Colors of constant
chroma at different value levels are connected by lines. Compare with Fig. 7-11a.

the range as one would expect from partitioning of the scale by a formula.
Here the first steps from gray were found to be in good agreement with the
average judgment of 35 observers (Kuehni, 2000b). As mentioned earlier, the
optimized linearizing powers are 0.84 for the green colors and 0.70 for the red
colors. They are 0.58 for the yellow colors and 0.33 for the blue colors. The
average increments in X and Z have the same ratio of 1 : 2.4 in OSA-UCS and
Munsell, but the average increment is 30% larger in OSA-UCS than in
Munsell. Figure 7-23 illustrates the Weber fractions for Z (comparable to Fig.
7-12) at six levels of lightness for colors on the j axis. They have a greater
regularity than those for the Munsell system for reasons mentioned.

7.3 THE SWEDISH NATURAL COLOR SYSTEM (NCS)

The designation “natural color system” has been used by Hering (Das
natürliche Farbsystem; Hering, 1905). To quote Hering: “For a systematic
grouping of colors the only thing that matters is color itself. Neither the qual-
itative (frequency) nor quantitative (amplitude) physical properties of the
radiations are relevant.”6 Thus Hering developed a system on perceptual basis
only. He postulated two achromatic percepts white and black and their ratio
W/B representing proportions. From this follows the expression for whiteness
302 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-22b OSA-UCS colors falling on the j axis at six levels of lightness L in the Z, Y diagram.
Achromatic colors fall on the diagonal, 10° observer and illuminant D65. Colors of constant
chroma at different value levels are connected by lines. Compare with Fig. 7-11b.

H = W/(W + B). Hering then described the development of a perceptually

equally spaced gray scale, based on the degree of similarity of a given step to
white and black. Most chromatic object colors, according to Hering, appear
veiled by the presence of whiteness, grayness, or blackness. “Chromatic colors
that do not obviously show such veiling I shall call unmasked chromatic colors.
. . .” Unmasked and veiled chromatic colors can be ordered according to hue
in a color circle. Hering defined four unique hues (Urfarben) and placed them
in such a manner on the hue circle that they divide it into its four quadrants.
Nearly one-half of this circle contains yellowish colors, nearly the other bluish,
and so for greenish and reddish colors. The colors in a quadrant can then be
defined by the relative amounts of the two adjacent unique hues in a way
similar to that described for achromatic colors. Hering placed the unmasked
(or “full,” Vollfarbe) color, black, and white at the corners of an equilateral tri-
angle into which all veiled colors of the hue of the unmasked color can be sys-
tematically arranged.
Hering believed that the clarity of chromatic colors could not be deter-
mined exactly; it can only be said of two colors which one is less veiled. “In
my view, every color that we actually see is more or less veiled. . . .” Any object
color perception can be described by its chromatic and achromatic compo-
 THE SWEDISH NATURAL COLOR SYSTEM (NCS) 303

Fig. 7-23 Weber fractions of Z for OSA-UCS colors falling on the j axis, at six levels of
lightness L, 10° observer and equal energy illuminant. Compare with Fig. 7-12.

nents: “It can be said that in each clearly veiled chromatic color both a
chromatic and a black-white component can be distinguished. . . .” Regarding
lightness/brightness of chromatic colors Hering said: “It is not such a simple
matter for chromatic colors, whose brightness or darkness is determined not
only by their black-white components, but also in part by their chromatic
components. . . . I have . . . ascribed an intrinsic brightness to yellow and red
and an intrinsic darkness to blue and green.” Hering’s equilateral constant hue
triangle therefore does not express anything about lightness or darkness of its
colors but only about the degree of veiling of its full color with white and/or
black.
As discussed in Chapter 2, Ostwald departed from Hering in attempting a
psychophysical definition while maintaining the geometrical form. Restricting
himself to colors that could be achieved with colorants, his full colors of neces-
sity fell short of Hering’s full colors. The form of his color solid is based on an
arrangement of Hering type constant hue triangles in a hue circle, resulting in
a double cone solid. But his hue circle is not based on unique hues but on
yellow, red, and blue primaries.
In the same chapter mention is made of the efforts of two Swedish
researchers, Johannson and Hesselgren, who tried to find a compromise
between Hering’s double cone and a color solid based in its third dimension
on a lightness scale. Johansson accepted hue, saturation, and value or lightness
as the three properties uniquely characterizing any color perception. He
304 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

suggested adding to these clearness or brilliancy and efficiency or vividness.

Hesselgren’s color atlas of 1952 relied on his idea of subjective color stan-
dardization. Regularity of spacing in that atlas was not considered to be
completely successful, and in 1964 the newly founded Swedish Color Center
Foundation began work toward a revision of the Hesselgren atlas by making
new psychophysical experiments. In 1966 Tonnquist reported on the deter-
mination of unique hues and a comparison between hue circles based on
unique hues (“symmetrical”) and those based on equality of perceived hue
differences. At the time the plan was to revise the atlas to be in principal
agreement with Johansson’s version of a natural color system. Eventually, a
decision was made to revert back to Hering’s original ideas and develop the
system accordingly (Hård et al., 1996).
In a pilot study forty subjects were divided into two groups, one of which
assessed a series of color samples according to the Hering principles by
reference to six samples representing the six elementary Hering colors. The
other group assessed them without reference to samples but based on their
internal concepts of the six elementary colors. While there were considerable
absolute differences in the average percentages between the two sets of
results, rank order showed a high level of correlation. The interobserver
variability was found to be low and comparable for both groups. Encouraged
by these results the Swedish Natural Color System (NCS) was developed on
the basis of judgments without reference samples.
From information provided one can estimate that in the basic experiment
about fifty observers visually assessed approximately 200 samples. Confidence
intervals at the 95% level of the means of the percentage judgments were
found to be less than 5 units on 100 unit scales each. Blackness was then
compared to luminous reflectance, chromaticness to the euclidean distance
from the neutral point in the CIE chromaticity diagram, and hue to the hue
angle in the same diagram. From interpolation on these psychophysical scales
some 900 samples were produced, and these were assessed in arrays by five to
six trained observers for smoothing purposes (“beauty test for acceptance”).7
The structure of the system is illustrated schematically in Fig. 7-24. It
consists of a double cone, formed in the atlas by forty equilateral constant hue
triangles. A gray scale forms the centerline connecting the two apices.

Fitting of Psychophysical Scales

A hyperbolic formula was fitted by least square method to the visually estab-
lished gray scale and was found to be in agreement with the Adams-Cobb
formula for a surround of Y = 56: w = Y(56 + 100)/(Y + 56). This was found to
be in reasonable agreement with the luminous reflectance of the light booth
interior used in the judgments (Y = 54), even though the luminous reflectance
of the immediate surround had Y = 78.
Other relationships to psychophysical scales were found to be much less
direct. Lines connecting samples with constant luminous reflectance were
 THE SWEDISH NATURAL COLOR SYSTEM (NCS) 305

Fig. 7-24 Schematic views of NCS. Left: double cone with circle of full colors. W: white; S:
black; F: location of a given color in the double cone. Right: Constant hue triangle with full color
C. The location of color F is determined by its blackness s and chromaticness c. From the two
values whiteness w can be derived. From Hård et al. (1996).

Fig. 7-25 Lines connecting colors of constant luminous reflectance for NCS hue R75B. YC
denotes the intrinsic blackness of full color C. From Hård et al. (1996).

plotted into constant hue NCS triangles and found to approximately intersect
at a point outside the triangle (Fig. 7-25). The position of this intersection point
varies widely by hue. Dominant wavelength of constant hue colors, as one
would expect, varies for most colors as a function of chromaticness and
lightness.
In the CIE chromaticity diagram or the a*, b* diagram NCS constant hue
lines are not straight, and constant chromaticness lines do not form circles,
pronouncedly so at higher chromaticness (Fig. 7-26). One of the reasons is
Hering’s arbitrary decision to value chromaticness of all full colors identically
at 100. In terms of chroma they vary significantly. Meaningful hue angle dif-
ferences between hue steps, such as presented in Fig. 5-25 for the Newhall data
set, cannot be calculated because of the very considerable variation in light-
ness of the full colors. The work of Billmeyer and Bencuya (1987) indicate the
considerable difference in apparent hue spacing of the Munsell Renotations
306 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-26 NCS colors of constant hue (radial lines) and constant chromaticness (ovoids) in
the CIELAB a*, b* diagram. Capital letters indicate the four NCS unique hues. From Derefeldt
and Hedin (1987).

compared to NCS. As discussed in Chapter 5, depending on what is true con-

stant chroma, the Renotations may contain some hue-chroma conflation. In
the case of NCS where constant hue steps at constant chromaticness involve
changes in hue, chroma, and lightness, hue steps likely involve a considerable
degree of conflation. The disagreement in apparent hue angle differences is
therefore not surprising. Indow has recently also begun to investigate the
psychological spacing of NCS (Indow, 2001). Of interest is the visual judgment
of the hue circle at constant blackness and chromaticness. A complete
circle was visually judged at s = 20 and c = 60 and a partial circle at s = 10
and c = 70. Based on summation of differences the segment angles in the
inner circle between the four unique hues of the system are R-Y 93°, Y-G 85°,
G-B 82°, and B-R 100°, These figures deviate considerably from those of
Table 4-1.
In a manner comparable to that shown above for the Munsell Renotations
and OSA-UCS, the colors falling near the axes in the X, Y, Z space are shown
in Fig. 7-27a and b. These figures demonstrate the significant difference
between attempted uniform color solids and NCS. While the systematic
arrangement of the samples is apparent it is of an entirely different nature
than that of roughly uniform solids. Developers of NCS have written computer
 THE SWEDISH NATURAL COLOR SYSTEM (NCS) 307

Fig. 7-27a NCS colors falling closest to the a opponent axis in the X, Y diagram. Achromatic
colors fall on the diagonal. Lines converging on the diagram origin connect colors of constant
NCS chromaticness, curved lines connect colors of constant blackness, 2° observer and equal
energy illuminant. Compare with Figs.7-11a and 7-22a.

software that converts, by iterative methods CIE colorimetric values, into NCS
values, and vice versa.

NCS Color Terminology and Atlas

Around the 40-step hue circle the hues (other than the unique hues) are iden-
tified in terms of their two constituting unique hues, for example, R70B, with
the meaning of a hue constituted of 30 parts redness and 70 parts blueness.
For veiled colors, according to Hering, the sum of the components of one or
two unique hues, whiteness and blackness, is always 100. Chromaticness c is
expressed as the percentage of chromatic components in the total color. The
full specification of an NCS color is expressed as in the following example:
4030-R70B, indicating a color of blackness s = 40, chromaticness c = 30, and
hue as defined above. The system and applications have been described by
Hård and Sivik (1981) and by Hård et al. (1996a, b).
The 1995 edition of the atlas contains 1750 color chips of 13 ¥ 15 mm size
arranged on forty pages (see Fig. 2-53). It is available in a matte and a glossy
308 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Fig. 7-27b NCS colors falling closest to the b opponent axis in the Z, Y diagram. Achromatic
colors fall on the diagonal. Lines converging on the diagram origin connect colors of constant
NCS chromaticness, curved lines connect colors of constant blackness, 2° observer and equal
energy illuminant. Compare with Figs. 7-11b and 7-22b.

version. Swedish Standard SS 01 91 03 contains the aim color specifications of

the atlas and many interpolated and extrapolated colors (16,800) in terms of
CIE tristimulus values and chromaticity coordinates for the CIE 2° observer
and illuminant C.

NCS System and NCS Atlas

The NCS system in its essence represents Hering’s idea of describing any
object color perception by its chromatic content and veiling with blackness
and/or whiteness. To this idea the developers of NCS added a color identifi-
cation methodology and the claim that any observer with normal color vision
can assess the chromatic, black and white contents of any object color with
good accuracy.
The atlas represents a curiously dissonant position. On the one hand, a
claim for the system is that it is independent of a specific reference frame and
that it can be used successfully in any kind of environmental situation. The
user, it is claimed, can make quite accurate judgments of the content of the six
primary color experiences of any perceived color in any situation. On the other
hand, colorimetric specification and quality control was required to meaning-
fully embody the system in the atlas.
 THE “FRAGILITY” OF COLOR ATLASES 309

The NCS atlas is a material example of the NCS system instantiated for a
particular surround and lighting condition: achromatic surround of Y = 56 and
CIE illuminant C, with the samples of a size to be, at normal viewing distance,
in agreement with the CIE 2° standard observer. Strictly speaking, for other
situations the atlas is less or not valid.

7.4 THE “FRAGILITY” OF COLOR ATLASES

Color atlases, such as those described above are fragile in a metaphorical sense
as well as in reality. The variability of observer responses to color stimuli and
the fact that less than ideal colorants are used to color atlas chips make color
atlases fragile in a metaphorical sense. The three described atlases all have
been prepared with the help of colorimetric tools. In the case of the Munsell
system the Renotations are defined in terms of colorimetric data from which
they can be instantiated by matching tristimulus values. The aim values have
been arrived at by plotting, smoothing, and interpolating results of visual
evaluations. In case of the OSA-UCS atlas a formula has been fitted to a set
of visual data, and that formula has been used to tile the space according to
the principle of interlocking cubo octahedra. The resulting L, j, g values have
been converted back to tristimulus values that are the aim points for the gen-
eration of atlas chips. A similar process of defining colors has taken place in
generating the NCS chips. A quality issue relatively easily definable and con-
trollable is that of the agreement between aim color and atlas color in terms
of colorimetric coordinates.
For more recent atlas editions an aim has been to select pigments that not
only are color fast but that result in colorations reasonably color constant
when the atlas is viewed in different phases of daylight or artificial daylight
sources. None of the three systems makes any but the most general claims in
this respect. The available selection of suitable pigments is not such that a sig-
nificant degree of demonstrated color constancy under different light sources
is possible. Atlases have been formulated for CIE daylight C or D65. There
are no artificial light sources that closely match these phases in spectral power
distribution. Artificial daylight sources are today usually based on fluorescent
lamps, and while they match the correlated color temperature of a particular
daylight phase, their spectral power distributions are often significantly dif-
ferent. So-called triband lamps (CIE illuminant F11) are particularly differ-
ent. As a result the appearance of the chips can change significantly, and the
original goal of uniformity or regularity of spacing is no longer met or met
less well.
The issue of surrounds has been discussed earlier, but a color atlas is valid
in its appearance strictly only for the surround color for which it has been
designed. Distortions in scaling take place if the surround is different in light-
ness or chromaticity from the design surround. Distortions due to surround
can be relatively large because of crispening effects.
310 MAJOR COLOR ORDER SYSTEMS AND THEIR PSYCHOPHYSICAL STRUCTURE

Since a numerical standard observer has been used in the design of atlases,
individuals with normal color vision viewing an atlas experience smaller or
greater distortions when viewing it, based on the difference between their
color vision apparatus and that implied for the standard observer, and likely
due to additional reasons. An idea of the magnitude of the changes in experi-
ence can be gleaned from the range of colors selected as unique hues by a
group of observers (see Chapter 1). In addition observers have varying
perceptions of equality of lightness, chroma, and hue steps to a degree that is
unknown. Different assessments of constant chroma circles by observer
groups in past extensive visual experiments, and different assessments of the
number of constant size hue steps between unique hues, as discussed in
Chapters 4 and 5, give an indication of the magnitude of these problems.
Aside from these conceptual fragilities there are also material fragilities. Of
particular importance here is the resistance of pigments to degradations of
various kind. Pigments, binders, and other materials are selected to make
modern atlases long lasting. However, much depends on the kinds and lengths
of exposures atlases suffer in use.

In this chapter three major color order systems have been compared, two
attempts at a uniform color solid and the third a Hering type color solid. The
considerable differences between the former two and the latter are evident
from views of the distribution of samples in the X, Y and Z, Y planes at the
system axes. Hue superimportance is implied in the Munsell system (as a result
of the unequal magnitude of hue and chroma scale intervals). It is nonuniform
in its euclidean form and non-euclidean when uniform. The system considers
neither the Helmholtz-Kohlrausch effect nor the surround lightness.The OSA-
UCS system, not arranged according to hue and chroma attributes (in fact, not
directly expressive of meaningful chromatic attributes), does not in its final
form indicate hue superimportance. However, analysis of the basis data clearly
shows the presence of this effect. Hue superimportance was suppressed in
the final formula used to tile color space according to this system to make
the space euclidean. As a result the system is not uniform. The suppression
explains the apparent difference in unit difference chromatic contours
between the Munsell system and suprathreshold small color difference data,
on the one hand (of generally oval shape), and OSA-UCS, on the other (cir-
cular). The circular unit contour is not in agreement with the perceptual facts
of the OSA-UCS basis data. OSA-UCS, however, is built under consideration
of the Helmholtz-Kohlrausch effect and of a specific achromatic surround. The
NCS solid has been developed based on Hering’s principles. Because the dis-
tances in the solid do not represent constant perceptual distances, there is no
problem fitting the solid into a double cone. With minor changes it could be
fitted into a sphere.We have arrived at an understanding that the three systems
in effect represent regular color solids, none of them being able to make a
strong claim for perceptual uniformity.
 Chapter 8
From Color-Matching
Error to Large
Color Differences

Color differences have been scaled from the level of color-matching error,
through threshold and industrially important suprathreshold small color dif-
ferences, large differences of the size in OSA-UCS, all the way to quadrant-
sized differences. In all cases considered here the difference judgments have
been made against a simple achromatic surround. In previous chapters we
have seen that there are systematic changes in terms of stimulus increments,
depending on the magnitude of the difference. In this chapter color differences
of varying sizes are compared in terms of stimulus increments to assess com-
monalities and divergences.

8.1 A COMMON BASIS FOR COMPARISON

It is of interest to compare such data both in L, M, S cone sensitivity space as

well as the CIE tristimulus space. Aside from scale differences the main dif-
ference between the two is that in case of the former comparison is made at
the point of interaction of light energy with cones, while in the latter case it is
made in a system that implicitly accounts for the reappearance of red at the
shortwave end of the spectrum and recognizes luminance and chromaticness
as important color attributes. Conversion to the L, M, S space has been cal-

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

311
312 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

culated using the Smith-Pokorny definition of cone sensitivities. All compar-

isons are made assuming an equal energy light source.
Historical color-matching error (CME) and suprathreshold small color dif-
ference data have usually been expressed in form of ellipses or ellipsoids in
the CIE chromaticity diagram or the x, y, Y space. More recently they typi-
cally have been expressed in the L*, a*, b* space or a cone contrast diagram.
Systematic studies of threshold differences and color-matching error have
started with the work of Wright (1941) and MacAdam (1942). MacAdam
continued the CME work together with Brown (1949), followed by Brown’s
work (1957) with twelve observers. Additional CME data were contributed by
Wyszecki and Fielder (1971a). These authors also reported on a different kind
of experiment where a chromatic difference was displayed and the observer
had to select a third color so that the chromatic differences between the three
colors were equally large (1971b). The variability in the setting of the third
color was then determined (CDM data). Richter reported in 1985 on an exten-
sive determination of threshold data. For the purposes of comparative analy-
sis it has been assumed that CME data represent object colors at a luminous
reflectance of Y = 30.
As mentioned earlier, a collection of suprathreshold small color difference
data based on several experiments using object color samples was assembled,
in parts modified, and normalized by Luo and Rigg in 1986 (L-R data). A
portion of these data, limited in luminous reflectance to the range Y = 25–35,
have been used in this analysis. Other sets of suprathreshold small color
difference data, such as the RIT-DuPont (R-D) data, were not considered
here because their range of luminous reflectance values is large, with a small
amount of data near Y = 30. Some of the data, including the R-D data,
were used to compare lightness scaling. There are several sets of published
large color difference data: Munsell, OSA-UCS, data by Wyszecki and Wright
(1965) developed in connection with a field trial for a color difference for-
mula, and the more recent Guan and Luo (1999) data. Color difference
formulas as representative approximate models of small and large color dif-
ference experiments, CIELAB and CIE 94, have been used for comparison
purposes.
Observer groups for the various experiments were obviously different, and
there were also many differences in observing conditions. Inconsistencies
between different data sets of the same type presumably are due to different
observer groups and/or different observation/test conditions. They also may
be due to effects so far not considered. Among these may be, as mentioned
earlier, the issue of significant variation in the apparent composition of the
lightness signal as well as the difference in sensitivity of yellow-blue and green-
red opponent systems found for different observers, the uncertainty of con-
stant chroma contours and constant hue differences (and very likely individual
variations thereof), and the resulting partial conflation of lightness, chroma,
and hue differences.
 CHROMATIC AND LIGHTNESS CRISPENING EFFECTS 313

8.2 CHROMATIC AND LIGHTNESS CRISPENING EFFECTS

In 1949 Le Grand analyzed the MacAdam ellipses in terms of increments in

two slightly different sets of three color fundamentals, R, G, and B. His results
indicated that for either set the experimental ellipses, when projected onto the
R, B plane, were nearly aligned with the axes of that plane, with the longer
semi axis aligned with the B axis and the shorter one with the R axis. When
plotting the increments representing the semi axis length as a function of the
value of the fundamental he found a function (similar to Fig. 8-1a) linear over
a large portion for B increments and a V-shaped one (similar to Fig. 8-1b) for
R increments. Boynton and Kambe reported comparable results in 1980 from
investigation of the size of incremental thresholds along lines of constant
values of fundamentals (note that these were not CME data). The increments
in B could be modeled as a constant fraction of B over a large range. Incre-
ments in R also confirmed the V-shape centered on the achromatic point,
discovered by LeGrand. Nagy et al. (1987) investigated CME data (MacAdam,
Brown-MacAdam, and Wyszecki-Fielder) and essentially confirmed the
Boynton and Kambe (and thereby the Le Grand) results. They fitted a
Fechnerian formula to the DS versus S function and a more complex formula
relating L and M and containing a component of S to the DL versus L func-
tion. The authors explained the V-shape of the latter function as “opponent
interactions between the two long-wavelengths cone mechanisms.” As an

Fig. 8-1a Increments in S as a function of S for the MacAdam ellipses (excluding ellipse 1).
Open circles represent points calculated by linear regression. They are connected by linear
regression lines.
314 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

Fig. 8-1b Increments in L as a function of L for the MacAdam ellipses (excluding ellipse 1).
Open circles represent points calculated by linear regression. They are connected by linear
regression lines.

aside, in a previous investigation Boynton et al. (1983) found that individual

observers differ in the relative sensitivity of their two opponent color systems
by more than a factor of 3.
Schönfelder, Kaneko, Takasaki, and others, as discussed in Chapter 4, have
found that threshold or unit larger difference increments determined against
an achromatic surround require the smallest stimulus increments/decrements
at the surround chromaticity. The required stimulus changes increase in
magnitude as the distance from the achromatic point increases. In case of
lightness differences the smallest change is also required at the lightness of
the surround.
Tristimulus values for the equal energy illuminant representing CME and
color difference data have been converted to Smith-Pokorny cone sensitivity
data and plotted as DS against S and DL against L (Kuehni, 2001d), and linear
models were fitted separately to both sides of the achromatic point. As men-
tioned, in the case of CME data, determined in visual colorimeters, the sim-
plifying assumption was made that they are object colors with a luminous
reflectance of Y = 30. In case of the Brown-MacAdam and Brown data with
variable luminosity, this was considered justified by MacAdam’s finding that
the size of CME ellipses varies by less than 20% as a function of luminosity
in a medium range of luminosities. Figures 8-1a and b illustrate the increment
functions for the MacAdam ellipses (minus ellipse 1).1 There is a fair amount
of variation and the regression lines do not intersect exactly at the achromatic
point. It is evident that both cases can be seen as representing V functions with
different angles of opening of V. Such a situation is what one would expect
 CHROMATIC AND LIGHTNESS CRISPENING EFFECTS 315

from the functioning of the chromatic crispening effect. Increments are, in

terms of stimulus, smallest at the surround chromaticity (achromatic point)
and grow, absolutely or at least relatively, in both directions away from that
point. The magnitude of the effect is different for L and S, in part due to the
fact that colors of constant luminous reflectance can differ much more in S
than in L or M, as we have seen earlier. (Because of the large degree of overlap
with L, the function for M is very similar to that of L and not shown.)
Comparable results have been obtained for all CME and suprathreshold
small color difference data. All slopes for L of colors where L is smaller than
that of the achromatic color are negative. In case of S it is negative for observer
GW in the Wyszecki-Fielder data. It is much less positive in all other cases
than that of the increments for colors with S larger than that of the neutral
point (except for the Brown-MacAdam data [observer WRJB] where there is
no apparent chromatic crispening effect involving S). A negative slope for S
less than that of the achromatic point is implied in the CIE 94 color difference
formula. Figure 8-2a and b illustrates the two functions calculated from a series
of CIE 94 ellipses in the a*, b* diagram with a constant total color difference.
The results indicate that the chromatic crispening effect is active, as one
might expect, in every set of CME, threshold and suprathreshold small color
difference data investigated. Its activity is implied in the SC function of the
CIE 94 formula and comparable functions in other color difference formulas.
The angle of opening of the V-shaped functions DL versus L and DS versus S
for various sets of data are given in Table 8-1. The angles are only compara-
ble within one type of function (the Richter data are not directly comparable
because the average Y is 16.5, rather than 30 as in all other data sets).

Fig. 8-2a Increments in S as a function of S for a series of ellipses according to the CIE94
formula in the a*, b* diagram.
316 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

Fig. 8-2b Increments in L as a function of L for a series of ellipses according to the CIE94
formula in the a*, b* diagram.

TABLE 8-1 Comparison of angles of opening of the V-shaped functions of DL versus

L and DS versus S in various data sets
Data set Angle, deg L Function Angle, deg S Function
MacAdam 110 142
Brown-MacAdam (observer WRJB) 147 180
Brown (weighted averages) 126 124
Wyszecki-Fielder CME (observer AR) 74 149
Wyszecki-Fielder CDM (observer GW) 81 141
Luo-Rigg ellipses (Y = 25–35 only) 110 116
Richter (Y = 16.5) 144 144
CIE 94 73 106

The results show that the V functions implied in CIE 94 are more sharply
angled than those of all experimental data sets. This indicates that the SC func-
tion of that formula adjusts for more than the chromatic crispening effect. The
implication is that unit small color difference contours in the a*, b* diagram
adjusted for chromatic crispening are oval in form. The additional effect of SC
is to convert the elongated contours to circles of equal size, as will be seen
later.
Using ellipses in the a*, b* diagram fitted to the Luo and Rigg and the RIT-
DuPont (R-D) data (Melgosa et al., 1994, 1997), one can determine the
average change in chromatic ellipse size at an approximately constant lumi-
nous reflectance. The average axis length (between L* = 56 and 64, n = 30)
increases by a factor 4.4 from chroma 0 to chroma 100 as a result of the chro-
matic crispening effect.
 CHROMATIC AND LIGHTNESS CRISPENING EFFECTS 317

Fig. 8-3a Regression lines of DS versus S of the selected Luo-Rigg ( Y = 25–35) and OSA-
UCS data normalized at the neutral point to those of the MacAdam data.

Fig. 8-3b Regression lines of DL versus L of the selected Luo-Rigg and OSA-UCS data
normalized at the neutral point to those of the MacAdam data.

In Fig. 8-3a and b the regression lines of the DL versus L and the DS versus
S functions of the L-R and the OSA-UCS data sets have been normalized at
the neutral point to those of the MacAdam data. Interestingly, for both cone
types the L increments are larger for colors both greenish and yellowish for
318 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

Fig. 8-4 DY/Y as a function of Y for the RIT-DuPont data. Open circles represent points cal-
culated by linear regression. They are connected by linear regression lines. Surround luminous
reflectance Y = 11.

the L-R data compared to the MacAdam data, and vice versa for the colors
both reddish and bluish.
A generally comparable situation applies to lightness crispening. We have
seen in Chapters 5 and 7 that lightness crispening is implicit in all experi-
mentally determined lightness scales. When plotting DL versus L of the lumi-
nous reflectance scale (or DY versus Y) we can expect to find a similar V
shaped form as in the case of chroma crispening, if the surround luminous
reflectance falls between the extremes of that of the test colors. Since in all
CME data the surround luminosity was lower than that of the least luminous
test field a V-shaped function is not expected in the data with variable lumi-
nosity, as indeed is not the case. In the R-D data, where the surround lumi-
nous reflectance was Y = 11, the effect is present but small, best illustrated by
plotting the Weber fraction of Y versus Y (Fig. 8-4). In terms of DL versus L
we can find the effect when plotting this function for OSA-UCS lightness that
has lightness crispening for Y = 30 built into the formula.The crispening effects
indicate that the Weber-Fechner law is not applicable to color differences
throughout the range of size where the crispening effects are applicable.

8.3 CHROMATIC CRISPENING FADES AS A FUNCTION OF THE SIZE

OF THE DIFFERENCE

When plotting DL versus L and DS versus S for colors of the Munsell and the
OSA-UCS systems along the axes in the a*, b* diagram, we find no chroma
crispening effect, and there is a continuous increase with a positive slope.
 CHROMATIC CRISPENING FADES AS A FUNCTION OF THE SIZE OF THE DIFFERENCE 319

The same applies for constant size chroma differences as calculated by the
CIELAB formula. When optimizing the divisor in SC for these color series, it
is found to be near 1 (Kuehni, 2001c). It is 1 when optimizing it to the hue and
chroma difference optimized OSA-UCS formula for the basis data of that set
(see Chapter 7). Guan and Luo (1999) optimized a modified CIE94 formula
to various sets of large color difference data, and found different optimal SC
and SH factors. They recommended for large differences a formula in which
the SC divisor is 1 + 0.016C* and the SH divisor is 1 (compared to SC divisor
1 + 0.045C* and SH divisor 1 + 0.015C* in CIE 94, GLAB, see chapter 6). An
explanation for these findings is that chromatic crispening fades as the size of
the chromatic difference increases. This is surprising and different from light-
ness where lightness crispening is present from color-matching error to large
differences.
In the absence of detailed experimental data one can estimate the relative
magnitude of the adjustment for the chromatic crispening effect in small color
difference perceptual data and the effect of converting ellipses to circles in the
SC divisor. If we assume the length of the major axis at the neutral point to be
1.5 (the average of 7 ellipses with C* < 2 is 1.55) and the increase in axis length
as a function of the chromatic crispening effect (see above) a factor 4.4, the
total value of the SC divisor should be 6.6 at metric chroma 100. It is found to
be only 5.5 in CIE94, a reasonable result given the variability in the visual
data.
It seems likely that this change in implied SC is located on a continuous
function, as estimated in Fig. 8-5. Systematic experiments are required to
clarify the shape of the function.

Fig. 8-5 Chromatic chrispening as represented by optimized SC divisor as a function of size

of CIELAB DE differences for three sets of data and linear estimate of the function relating the
two. The boxes represent the approximate ranges of differences in the data sets.
320 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

8.4 SIZE AND RATIO OF UNIT INCREMENTS

Calculation of the linear regression lines of incremental data allows com-

parison of the size of the unit increments. Because of the different magnitude
of the chromatic crispening effect in different data sets meaningful comparison
is only possible of the implied first step from the achromatic point. These are
compared in Table 8-2 in the direction of higher L respectively S values.
Several observations and conclusions can be drawn from the data in this
table:

1. MacAdam and Brown-MacAdam data are very similar.

2. Observers GW and AR differ significantly in their sensitivity to sub
threshold L differences.
3. The Brown average data differ significantly, either as a result of the
weighted averaging process or for other reasons, from the other CME
data.
4. The color difference matching error is elevated by approximately a
factor 2 from the color-matching error.
5. The Richter threshold data indicate a higher L increment compared
to CME data but a lower S increment, resulting in a significantly lower
ratio.
6. The selected L-R data also have a lower ratio and even more pro-
nouncedly the CIE94 and the CIELAB formulas.

TABLE 8-2 Comparison of unit L and S increments for the first step from the neutral
color toward higher L and S values and their ratio for various data sets based on
linear regressions, Y = 30
Data set L Increment S Increment Ratio
CME data
MacAdam 0.0168 0.40 23.8
Brown-MacAdam 0.0232 0.60 25.8
Wyszecki-Fielder/GW 0.0191 0.72 37.9
Wyszecki-Fielder/AR 0.0300 0.78 26.0
Brown 0.0410 0.21 5.1
Wyszecki-Fielder/CDM 0.076 1.53 19.7
Threshold and small color difference data
Richter (extrapolated to Y = 30) 0.036 0.38 10.6
Luo-Rigg data 0.045 0.63 14.0
Large difference data
Munsell 0.373 6.00 16.1
OSA-UCS 0.529 8.22 15.5
Color difference formulas
CIE94/DE = 1.0 0.043 0.31 7.2
CIELAB/DE = 1.0 0.103 0.66 6.4
 SIZE AND RATIO OF UNIT INCREMENTS 321

TABLE 8-3 Comparison of unit increment values of X and Z for the first step from
neutral for selected data, based on regression calculation at Y = 30
Data set X Increment Z Increment Ratio
CME data average (excluding Brown) 0.136 0.625 4.6
Luo and Rigg 0.279 0.629 2.3
Color difference matching data 0.471 1.50 3.2
Munsell data 2.35 6.03 2.5
OSA-UCS 3.23 8.22 2.5
CIELAB formula, DE = 1.0 0.639 0.658 1.0
CIE94 formula, DE = 1.0 0.265 0.310 1.2

7. Munsell and OSA-UCS have similar ratios, and the OSA-UCS steps are
approximately 1.4 times larger than two Munsell chroma steps, as we
have seen in Chapter 7.

The average L increment of the CME data (excluding the Brown data) is
0.022, the S increment 0.625, for a ratio of 28.4. For the L-R data selection, the
ratio is 14.8. It is evident that a significant change is happening between imper-
ceptible color-matching errors and small color differences: while S increments
are nearly the same it is L increments that become larger in small color dif-
ferences by approximately a factor of 2. The ratio remains essentially the same
when the differences are large. The color difference matching error has a ratio
smaller than that of color-matching error but larger than that of color differ-
ences. The Richter threshold data with a ratio of 10.6 are an exception to the
picture, due to a very low S increment.
When data are viewed, in the CIE tristimulus system a more familiar picture
emerges. Table 8-3 contains a selection of the data of Table 8-2 expressed in
terms of X and Z (the latter identical to S).
Small and large color difference data have comparable ratios of approxi-
mately 2.5. The CIELAB formula, by definition, has a ratio of 1, due to the
factor of 2.5 between the multipliers of a* and b* (500/200). CIE94, due to the
effect of SC, has a slightly larger ratio. Unsurprisingly, the key difference
remains the same: while in the CME and small color difference data the Z
increments are the same, the X increment increases by an approximate factor
of 2 in the small and large color difference data, compared to CME data. It
should be recalled that the multiplier balancing the linear opponent color
functions a and b is 2.3 to 2.4, depending on the standard observer. The impli-
cation is that color difference perception regards the balanced a and b systems
as equivalent.
On the other hand, color-matching errors appear based only on the sensi-
tivity limits of the cones. (Here the findings of Boynton and Kambe, 1980, of
thresholds along constant S and constant L/2M lines are not in agreement,
since they largely duplicated, if with larger Weber fractions, MacAdam’s
results.) Color-matching error data and color difference data are, it appears,
322 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

incommensurable, and geodesics based on color-matching error data do not

predict geodesics of color difference data. The Richter threshold data, as one
would expect, fit into this picture as indicated by an, albeit low, ratio of 1.7.

8.5 DIRECTION OF UNIT CHROMATIC CONTOURS IN THE L, M, S

AND X, Y, Z SPACES

The difference between CME and color difference data is further clarified
when comparing the shape of their chromatic contours in L, M, S and X, Y,
Z spaces. For this purpose the intersection points of the major and minor axes
of the unit contours are translated into these spaces and views in certain direc-
tions are created. The exercise is limited to representative data: the MacAdam
ellipses and the Wyszecki-Fielder (observer GW) data as representative for
CME data and the Luo-Rigg data for small suprathreshold color differences.
Figures 8-6 to 8-8 illustrate the ellipses, each represented by five points (the
endpoints of the long and short ellipse axes in the a*, b* diagram and the
centerpoint), for the three cases in the L, S view. A number of observations

Fig. 8-6 MacAdam ellipses represented by the center point and the four points of the ellipse
intersections with the major and minor axes, in the L, S cone response diagram, equal energy
illuminant.
 DIRECTION OF UNIT CHROMATIC CONTOURS IN THE L, M, S AND X, Y, Z SPACES 323

Fig. 8-7 Selected Luo-Rigg ellipses represented by the center point and the four points of the
ellipse intersections with the major and minor axes, in the L, S cone response diagram, equal
energy illuminant.

TABLE 8-4 Angles of major ellipse axis against abscissa in the L, S and X, Z
diagrams
Data set Number of Angles Mean Angle, Range, deg COV, %
deg
<90 >90
L, S diagram
MacAdam 16 9 89.8 86.2–96.0 0.1
Wyszecki-Fielder/GW 5 23 92.6 87.0–101.0 3.9
Wyszecki-Fielder/AR 8 20 91.0 87.9–99.9 2.1
Luo-Rigg (Y = 25–35) 5 26 93.1 86.1–99.4 3.5
X, Z diagram
MacAdam 22 3 77.4 57.4–115.0 16.8
Wyszecki-Fielder/AR 21 7 87.9 65.1–136.5 19.8
Luo-Rigg 9 22 96.8 56.9–130.6 19.2

can be made from the results. The MacAdam ellipses are well aligned with the
axes of the diagram; that is, they are well described by differences in L and S.
Table 8-4 contains data concerning the angles against the abscissa of the major
ellipse axis for some data sets in both the L, S and the X, Z diagrams.
The essentially vertical alignment of the MacAdam ellipses in the L, S
324 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

Fig. 8-8 Wyszecki-Fielder (observer GW) ellipses represented by the center point and the four
points of the ellipse intersections with the major and minor axes, in the L, S cone response
diagram, equal energy illuminant.

diagram is confirmed by the statistics. Both Wyszecki-Fielder observers show

a bias toward angles larger than 90°, primarily due to a counterclockwise rota-
tion of contours in quadrant 4 (yellowish-reddish colors; Fig. 8-8). This effect
is stronger in the GW data. The bias in the L-R data is comparable to that of
observer GW.
Because of the rotation in space and the rescaling of the X compared to the
L axis, the picture looks different in the X, Z diagram, as shown in Figs. 8-9 and
8-10. Here most MacAdam ellipses have a strong bias toward a smaller angle,
and the variability in angles has increased significantly. The AR data have a
similar tendency and the variability in angles is even larger. The L-R data con-
tinue to have a bias toward angles larger than 90°, also with a high variability.
Here the tendency of many ellipses is to point in the direction of the neutral
point (at X, Z = 30).The X, Z diagram is a not a normalized version of an oppo-
nent color diagram, and in a normalized diagram the tendency is even clearer.
The implication is that these ellipses are aligned along constant hue lines
and represent the larger increment required for a unit chroma difference
compared to a unit hue difference of the same perceptual size. This tendency
is also apparent in one quadrant for observer GW, less so for observer AR. In
 THE PARADOX OF HUE DIFFERENCES 325

Fig. 8-9 MacAdam ellipses represented by the center point and the four points of the ellipse
intersections with the major and minor axes, in the X, Z diagram, equal energy illuminant.

the Munsell system the contours, as we have seen in Chapter 7, are uniformly
aligned along radial lines. In the X, Z diagram this is more or less the same.
The conclusions one can draw is that under the conditions of the MacAdam
experiment and/or as a result of the vision properties of that observer the
color-matching error is caused by limitations in cone sensitivity only. In the
conditions of the Wyszecki-Fielder experiment yellowish-reddish colors of
quadrant 4 point toward the neutral point, but the ratio between the two axes
remains average for CME data. An opponent color system appears active in
case of the L-R suprathreshold data. Here in addition the L, respectively X,
unit increment is increased compared to the S or Z increment in accordance
with global requirements for uniformity.

8.6 THE PARADOX OF HUE DIFFERENCES

Pages of the Munsell or NCS atlases contain by design colors of constant hue
at different levels of lightness and chromaticness. One might assume that the
hue differences between corresponding colors on two adjacent pages are all of
identical size, since they all bear the same respective hue names (i.e., that hue
326 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

Fig. 8-10 Selected Luo-Rigg ellipses represented by the center point and the four points
of the ellipse intersections with the major and minor axes, in the X, Z diagram, equal energy
illuminant.

differences are independent of chroma and lightness differences). Diagrams

such as the Munsell psychological chromatic diagram or the CIELAB diagram
show a much different picture. The CIELAB formula predicts that the hue dif-
ference between two colors of identical chroma and lightness, and differing
only in hue, varies by a factor of 10 between metric chroma of 10 and 100. A
comparable result is implicit in the Nickerson Index of Fading. The paradox is
that two pairs of colors with the same hue names at two different chroma levels
with the same hue angle difference have a calculated Nickerson or CIELAB
hue difference differing by a factor of 10. (It should be noted that in the
CIELAB formula hue superimportance is not considered and the ratio
between unit hue and chroma difference does not conform to the visual ratio.)
If this is valid, the implication is that at metric chroma 100 there should be 10
times more distinguishable hue differences compared to metric chroma 10.
Hue, of course, fades at the achromatic point and hue differences disappear
also. The author is not aware of any study other than Nickerson’s where the
size of perceived hue differences between sets of Munsell colors of constant
hue as a function of chroma and lightness was investigated explicitly. The
Munsell BOC illustrates all forty hues down to chroma 2 and half of them to
 THE PARADOX OF HUE DIFFERENCES 327

Fig. 8-11 Lines connecting two hypothetical series ( m and n) of colors of constant hue in a
perfect hue circle. Line o represents the colors of line n after application of the SH hue differ-
ence weight in the CIE94 color difference formula, normalized at metric chroma 25.

chroma 1. The only (implied) quantitative psychophysical data we have in

this matter are those used to determine the SH factor in the CIE94 and similar
formulas. It should be recalled that it is based on fitting a formula to elliptical
contours in the a*, b* diagram so that they become circles of equal size. No
hue difference judgments have been made in its support.
When plotting the hue differences implicit in CIE94 and similar formulas
as a function of metric chroma (Fig. 8-11) we find a different picture applica-
ble at the small color difference level. Figure 8-11 schematically illustrated two
series of colors each of constant hue but varying in chroma. The hue angle
between members of the two series is constant. Line o shows the effect of the
SH factor, normalized at C* = 25. The result is in better agreement with infor-
mal evaluations of the Munsell hue circle: perceived hue difference between
constant hue pairs fades to zero at the achromatic point but more gradually
than the CIELAB formula implies. If the hue difference between line m and
the normalizing point on line o is 1.0, it is 1.4 at chroma 10 and 0.30 at chroma
100. That is to say, colors on neighboring equal hue lines differ nonlinearly in
328 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

Fig. 8-12 Uniform chromatic Riemannian plane constructed from segments between lines m
and o in Fig. 8–11.

hue difference, from metric chroma 10 to 100 approximately by a factor 4.6

rather than the factor 10 implicit in the euclidean chromatic diagram. The SH
factor should be verified with direct hue difference judgments.
There are two important conclusions to be drawn from this result:

1. Constant hue angle difference between colors of two hues does not imply
constant perceived hue difference.
2. As a consequence a color space based on SH cannot be euclidean because
uniform slices created by SH do not add up to form a flat circular plane.
When forming a complete hue/chroma plane from segments between
lines m and o of Fig. 8-11, a Riemannian plane (Fig. 8-12) is obtained.

8.7 UNIT DIFFERENCE CONTOURS AROUND THE HUE CIRCLE

Hue Angle versus Ellipse Angle

Color difference formulas derived from CIELAB imply that (with exception
of colors of near unique blue hue) colors of constant hue lie on radial lines in
the a*, b* diagram. While this is reasonably well the case for constant hue lines
of the Munsell system, it is less so for those of NCS. As a first approximation
we can assume, however, that this rule applies. The question arises if the con-
jecture that hue and chroma difference perception mechanisms shape unit
difference contours of color difference data is well supported by data at the
small color difference level. To assess this matter, 139 ellipses fitted to L-R
(Melgosa et al., 1994) as well as R-D data (Melgosa et al., 1997) were investi-
gated (11 with metric chroma <5 were excluded). The correlation coefficient
between the CIELAB hue angle of the center color and the ellipse angle Q
 UNIT DIFFERENCE CONTOURS AROUND THE HUE CIRCLE 329

Fig. 8-13 Scatter diagram of angle of major ellipse axis (ellipse angle) versus hue angle for
the combined set of Luo and Rigg and RIT-DuPont ellipses with linear regression line.

was calculated as 0.95. As the scatter diagram of Fig. 8-13 indicates, significant
discrepancies appear to be random. When investigating the R-D data sepa-
rately (n = 16), the correlation coefficient is found to be 0.88. After deleting
two outliers that indicate longer axes in the hue direction than the chroma
direction, the correlation coefficient increases to 0.98. Angles of the 11 near-
neutral ellipses for the complete data set are found to range from 53 to 136.
While it is evident that CIELAB is not a good basis model for a uniform
chromatic diagram and there is the fact that transformation of ellipses fitted
in the CIE chromaticity diagram into ellipses in a*, b* involves a certain
amount of error (Melgosa et al., 1994), the surprisingly high correlation coef-
ficient appears to provide solid support for the conjecture of unit contours
being aligned with constant hue lines. The conjecture is found to apply also to
unit contours in the Munsell system and, as seen in Chapter 7, in OSA-UCS.
A tendency in this direction was also found in quadrant 4 of the Wyszecki-
Fielder color-matching error data. The alignment of fitted small color differ-
ence contours in the chromatic diagram has resulted in color difference
calculation being performed in a polar coordinate rather than a rectangular
system.
Additional data have in recent years been provided by J. Krauskopf and K.
R. Gegenfurtner (K-G, 1992) and by M. J. Sankeralli and K. T. Mullen (1999).
K-G determined visual thresholds around sixteen colors placed equidistant
330 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

Fig. 8-14 Visual thresholds around sixteen colors placed equidistant from the white point in
the center of the cone opponent diagram. From Krauskopf and Gegenfurtner (1992).

from the white point in a cone threshold diagram (see Fig. 8-14). In this
diagram the scales of the two axes have been adjusted so that a circular
contour is formed by the thresholds surrounding the white point. Most of the
resulting fitted threshold contours are ellipses more or less pointing toward
the achromatic point of the diagram. The fact that they do not closely do so
in all cases is an aspect of the controversy concerning the number of hue detec-
tion mechanisms in the human visual system (as briefly discussed in Chapter
5). Sankeralli and Mullen (1999) determined unit chromatic contours at the
45° and 135° angles of a normalized cone based diagram with the axes L - M
and S - (L + M)/2 in the observer’s personal isoluminant plane. The more or
less elliptical contours are aligned well with the corresponding hue angles.

Ellipse Shape, Size, and Relation to Hue Angle, Metric Chroma, and
Metric Lightness
In Chapter 7 it was shown that the ratio of contour axis length for the Munsell
colors at moderate chroma has been determined as approximately 2 : 1 (2.8 : 1
at JND level). On average, a ratio of about 2 : 1 applies to the OSA-UCS data
(see Section 7.2). Based on a regression of axis length versus metric chroma
of the combined Luo and Rigg and RIT-DuPont set, the average major and
minor axis lengths are found to be a = 1.7, b = 0.80 at C* = 0 and a = 4.1 and
b = 1.92 at C* = 100 (ratio of 2.14 : 1). For seven near-neutral colors the ratio
is found to be 1.55 : 1. The value of 2.14 : 1 is less than that found by Bellamy
and Newhall. The average ratio of the Krauskopf and Gegenfurtner contours
is found to be 1.7 : 1. The values for the Sankeralli and Mullen contours differ
 UNIT DIFFERENCE CONTOURS AROUND THE HUE CIRCLE 331

TABLE 8-5 Correlation coefficients for the relationship between ellipse area and
metric lightness L*, Luo and Rigg and RIT-DuPont data
Metric Chroma C* Correlation Coefficient Number of Samples
10–20 0.06 21
20–30 -0.19 25
30–40 -0.64 27
31.00–34.99 -0.63 15
40–50 -0.10 21
60–70 0.19 11
I calculated these rusults from the L–R and R–D data.

by observer, with a ratio of 2.4 : 1 for one and 1.9 : 1 for the other. Undoubt-
edly, the experimental conditions also affect the ratio.
When determining the relationship between the length of the major axis
and metric chroma for the L-R and R-D data, a correlation coefficient of 0.77
is found (n = 150), indicating good correlation. An only slightly lower value is
found when comparing ellipse area against metric chroma (0.74). This level of
correlation is another expression of the activity of the chromatic crispening
effect. On the other hand, no correlation was found when comparing ellipse
area and hue angle. In the limited metric chroma range of C* = 30 - 40
(n = 27) the correlation coefficient is -0.01.
An unexpected result was obtained when comparing ellipse area against
metric lightness L* as a function of metric chroma. The results are found in
Table 8-5. They indicate a considerable negative correlation between lightness
and ellipse area for metric chroma values between approximately 25 and 45.
That is, in this chroma range the ellipses become smaller as lightness increases.
Based on the regression the average ellipse area in the metric chroma range
of 30 to 40 is 18.1 at metric lightness 20 and 8.7 at metric lightness 80, a ratio
of approximately 2 : 1. It is not evident what the cause of this finding is, and it
requires further investigation and formula fitting.
As discussed in Chapter 4, there is little doubt that lightness and hue
are fundamental color attributes. Chroma (contrast) is the necessary third
attribute to represent all possible object colors systematically. Regardless of
the mechanism we assume for the generation of hue (two process or multi-
process) it seems to be a matter of fact that in terms of a psychophysical
presentation, we are more sensitive to stimulus increments if they signal
hue changes than if they signal chroma changes. In a psychological or psy-
chophysical diagram the outcome is an elongated unit contour. In practical
terms, there appear to be two independent systems: one that assesses changes
in the ratio of two opponent color signals (assuming a two-process hue
detection system) and the other changes in the size of the vector sum of the
opponent color system (indicative of contrast). Both are affected by surround.
The two seemingly operate independently of each other and are not connected
in a euclidean sense. A euclidean space can only be achieved (and only for
332 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

small color differences) with the help of euclidization operations such as the
one by Thomsen.
In Chapter 5 we observed that there are different numbers of constant size
hue differences between unique hues. As a result, in a chromatic diagram in
which constant hue differences occupy equal hue angles (e.g., the Munsell
system), unique hues do not fall on the diagram axes. On the other hand, in a
diagram where the unique hues fall on the axes, equal hue angle differences
do not indicate equal hue differences. To understand this situation, we require
knowledge of the mechanism resulting in unique hues and of the hue
difference mechanism.

8.8 GLOBAL DIFFERENCES

Differences of the Munsell or OSA-UCS step size are not the largest dif-
ferences that can be compared in a color space. It is possible to think of more
global kinds of differences by posing questions like:

1. Are the hue differences between the four unique hues the same?
2. Are the differences between colors falling on axes (however chosen) in
the Munsell psychological diagram and the central gray of the same mag-
nitude, and is gray in fact exactly in the middle between such colors?

The first question has already been answered in terms of the varying number
of hue differences between the unique hues. Questions of the second kind were
of interest to Judd and discussed by him in his article Ideal color space (1969).
Based on his experiments with Munsell chips, he concluded that the superim-
portance of hue gradually fades as hue differences become very large.
Of Judd’s observers a majority decided that the shortest path between
opposing axis colors was not through gray but through a desaturated, hued
color halfway between the opposing colors being compared. Judd worked out
a function that accomplishes this fade for the modified Godlove difference
formula (equation 4-2). But he indicated that there appears to be no geomet-
rical model corresponding to his formulation, “at least, I have not been able
to think of one.” He concluded that also from this viewpoint ideal color space
was impossible. Additional complications appear to arise when considering
very large color differences.

8.9 HOW FUNDAMENTAL ARE THE VARIOUS KINDS OF DATA?

A goal of vision science is to uncover fundamental color vision processes. For

this reason experimental conditions tend to be rigorously controlled to expose
only one assumed process at a time to experimental variability. Exposures to
 HOW FUNDAMENTAL ARE THE VARIOUS KINDS OF DATA? 333

test stimuli tend to be very short to minimize changes in adaptation due to the
test stimulus.
Color-scaling experiments, on the other hand, usually involve unlimited
exposure times. Experimental conditions usually varied between different
experiments and the degree of adaptation (in experiments involving small or
large suprathreshold experiments) were not considered nor decisively con-
trolled. As a result it is unlikely that fundamental processes have been deter-
mined but rather that the results represent overlapping fundamental processes
(if it is possible to identify fundamental processes in this manner), however,
often representative of practice in visual color quality control and of more
natural viewing conditions.
In natural viewing conditions adaptation tends to change continuously as a
result of what is being viewed, and the observer controls exposure times. If I
harvest ripe strawberries in a strawberry patch, as I change my gaze, the visual
system is adapted in various degrees to, among other things, combinations of
the brownish-grayish color of soil, the green leaves of the strawberry plants,
and, if there are plenty of ripe strawberries, the redness of the fruit. It is known
that some adaptation processes proceed very rapidly in time while others are
slower. Perceived color differences between two adjoining strawberries are a
result of the total momentary adaptation situation.
It remains to be seen how large the differences are between the two kinds
of observation conditions. In addition the observer groups in different
experiments most always are different. Variations in the results of different
observers are well known. They can be due to variations in their cone response
functions or other components of their color vision system. The degree of
linear transformability of individual color-matching functions remains to be
determined as well if meaningful transformable means can be calculated
from individual results. As discussed earlier, there is the possibility that certain
judgments may be influenced by evolutionary experiences.
It is important to keep in mind that the following conclusions are valid only
for a general viewing condition where the surround is achromatic and in
luminance/lightness somewhere between the upper and lower extremes of the
targets used in testing. At the color-matching error level we have found
variation among observers and, possibly, experimental conditions. The
MacAdam data are well represented by elongated rectangles in the L, S
diagram. Both Wyszecki-Fielder observers here investigated (GW and AR)
produced ellipses pointing already to some degree in the direction of the
neutral point, most strongly so in the fourth quadrant (this applies also to
the third observer). The impression is created that under the conditions of the
MacAdam experiment, or at least for that observer, CME is not influenced by
specific properties of the opponent color system but is based only on cone
response limitations. Under the conditions of the Wyszecki-Fielder CME
experiment, rotation of unit ellipses in the direction of the neutral point begins
to be noticeable particularly in one quadrant. The same applies to the color
difference matching results. At the level of thresholds the opponent color
334 FROM COLOR-MATCHING ERROR TO LARGE COLOR DIFFERENCES

system appears fully engaged (change in X to Z ratio, general alignment of

unit contours in the direction of the neutral point) and remains so through
large color differences.
It is not obvious what the cause of the rotation of a specific portion of the
contours in the Wyszecki-Fielder CME and CDM data is. The MacAdam and
Brown-MacAdam experiments employed monocular viewing. In both cases
the surround was neutral. In the case of the MacAdam experiment, it was at
half the luminance of the test fields; in case of Brown-MacAdam, it was dark.
Ellipse rotation toward the neutral point is also noticeable in the five most
highly saturated yellowish red colors of observer WRJB. In the Brown exper-
iment a binocular colorimeter was used with a neutral surround slightly less
luminous than the least luminous of the test fields. Ellipse rotation here is
evident for the two most highly saturated yellowish red colors. The Wyszecki-
Fielder CME experiments used two fields of a seven-field binocular
colorimeter. The neutral surround was at half the luminance of the test field.
As is mentioned in Wyszecki-Stiles, “. . . both eyes . . . could wander over the
field with no strict fixation. . . . These conditions . . . approximated most
closely those of ordinary viewing.” The surrounds in all cases were large
compared to the test fields. As mentioned, ellipse rotation is evident in the
yellow-to-red segment for all three observers.

Data analysis in this chapter has been pursued based on a conventional view
of color vision. From the results we can draw a number of conclusions regard-
ing the various divisions of a common psychophysical color space (e.g.,
CIELAB), based on the magnitude of differences involved. Before going into
details it should be mentioned again that the experimental basis at the various
levels tends to be different. Color-matching errors and thresholds are usually
determined using visual colorimeters. Small suprathreshold, medium, and
large differences have usually been determined using color samples of some
kind viewed in varied surroundings or, more frequently, in a light booth,
usually in an achromatic surround.
The results of observer PGN in the MacAdam experiment indicate that hue
discrimination was not explicitly active but that they are strictly a result of the
sensitivity limitations of the three cone types. For both observers, in the
Wyszecki-Fielder experiment, hue discrimination appears partially engaged (a
number of ellipses point to the origin), but not all other aspects of the oppo-
nent color system (the average ratio between L and S increments is the same
as in the MacAdam experiment and twice as large as in the Luo-Rigg and large
color difference data). This implies engagement of different aspects of the
system under different conditions. Lightness and chromatic crispening effects
are engaged for CME data.
Based on the Richter data, at the threshold level all identified components
are active. Suprathreshold small color difference unit ellipses, as we have seen,
have a quite strong tendency to point to the origin. This effect is geometrically
enhanced in the X, Z diagram, and even more if the X axis scaling is expanded
 HOW FUNDAMENTAL ARE THE VARIOUS KINDS OF DATA? 335

by a factor 2.3 for numerical balance with Z. Power modulation, lightness, and
chromatic crispening apply at this level. The L or X increments have doubled
compared to CME data, indicating increments in L or X that are not guided
by cone sensitivity but by another mechanism, an opponent color system.
In differences of Munsell and OSA-UCS magnitude the opponent color
system and hue discrimination are active, explicitly so in case of Munsell and
implicitly in case of OSA-UCS, as we have seen. Lightness crispening is active
as seen in the pre-1943 Munsell data and as recognized by inclusion of the
effect in OSA-UCS. Chromatic crispening, however, has now faded. The result
indicates that the processes guiding uniform tiling of color space are different
at different levels of size of differences.
 Chapter 9
Conclusions and Outlook

In our travels through psychological and psychophysical color spaces it was

shown that a uniform color space can only be defined for a narrow set of con-
ditions and that there is a multiplicity of uniform spaces, each representative
of a specific observation situation. It also has become clear that there are an
infinite number of regular color spaces in which colors are ordered in sys-
tematic sequence (and in the same ordinal order as the perceptual space) but
are not uniform. Color solids can fit into any desired euclidean space form if
the concept of uniformity is absent. This text is limited to the case of achro-
matic surrounds, a situation that is not expressive of most natural viewing con-
ditions but represents a useful elementaristic experimental simplification. It is
also in agreement with usual industrial observation conditions for the purpose
of color quality control. On this journey we have learned that we have only a
fraction of the information required to build accurate models of human color
vision that speak to the problem of uniform color space.
There are various uses of a color solid, and they have resulted in several
different major kinds of spaces. There is the uniform color space, classically
defined by Judd, in which equal geometrical distances in any direction (given
standard viewing conditions) represent equal perceived differences. There is
Hering’s “natural color system” in which colors are ordered by their content
of unique hues as well as blackness and whiteness. Here lightness is not an
attribute and chromaticness is uniform for all maximal colors. There are
nonuniform color spaces used in communication and color management, such
as the RGB, HSB, or CMYK spaces. Finally there are a number of psy-

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.

336
 WHAT ARE COLOR SPACES AND HOW CAN THEY BE JUSTIFIED? 337

chophysical color space models that order colors and generally have no claims
of uniformity, such as the CIE tristimulus space, the x, y, Y space, the Luther-
Nyberg space, and many others. For those without claim to uniformity, the
general designation regular color space has been proposed. In addition there
are spectral spaces derived from mathematical dimensionality reduction of
spectral functions of color stimuli not involving cone or color-matching
functions.
For a period of some 150 years there was belief that all object color per-
ceptions, uniformly spaced, could fit into a simple geometrical solid. Munsell
was the first to show that if uniformity of differences is a goal this belief is mis-
placed. Since then it has become evident that a Munsell level uniform color
solid cannot fit into a euclidean space.
The following is a discussion of a number of significant issues related to
color spaces and solids with stated or proposed answers (depending on the
strength of current evidence).

9.1 WHAT ARE COLOR SPACES AND HOW CAN

THEY BE JUSTIFIED?

Color spaces and color solids represent attempts to show a multitude of pos-
sible color perceptions in a systematic fashion. This is an effort that faces
several philosophical and scientific problems. We do not know how we per-
ceive colors as part of our consciousness. From an extreme point of view (see
Saunders and van Brakel 2002) this dooms the entire enterprise. We can all
agree that there is no generally valid direct relationship between spectral
power entering our eyes, and thereby between reflectance functions of objects,
and our experiences of color. Experimental work in color scaling under ele-
mentary conditions and work in color technology indicate, however, that for
such conditions there is a reasonably solid relationship between members of
a class of metamers and the resulting color experience as judged by an average
color normal observer. Without it, colorant formulation and color technolo-
gies of various kinds (color photography, television, color printing, etc.) would
not work as well as they do. Is viewing color fields under elementaristic con-
ditions a true reflection of natural color experiences? The answer, obviously,
is no. However, the evolutionary development of human color vision should
not be confused with its current average operation. Arguments can be made
that (in industrial countries) we experience color today more often in unnat-
ural as well as quasi elementaristic conditions (looking at a video screen or
color television) than in natural conditions (except when on vacation). Does
it mean our color experiences are degraded? Does the color vision system of
young children now develop differently from that of our grandparents? We do
not know but it is unlikely to be so.
The hue sequence of the generally accepted hue circle derives its justifica-
tion from the spectrum and the mixture in various ratios of narrow band
338 CONCLUSIONS AND OUTLOOK

energy from the two ends of the spectrum. For the reason that color experi-
ence is private we can only assume that one person experiences the colors of
spectral lights approximately the same way the next one does (assuming they
do not have impaired color vision). That photometry can produce an ordinal
scale of gray objects viewed under elementaristic conditions is without doubt.
Perceived brightness of chromatic objects, as was shown, is not in agreement
with photometric results, for reasons that we do not know. Speculative models
can, however, result in a high level of agreement with average visual data.
Quantitative relationship at the ordinal level for elementary conditions also
exists quite clearly between spectral power distributions of metameric sets
and the resulting perceived saturation or chroma. All these relationships have
high rank order correlations. Such information has provided the basis for
simplistic color models attempting to relate under elementaristic conditions
physical data of lights or materials with the resulting average perceived color.
For the models to be quantitatively useful they must have a higher degree of
complexity. There is no doubt, and the history of chapters 2 and 6 clearly indi-
cates it, that often (in an absolute sense always) large houses of cards are
erected in the building of these models. But as the history of science shows
this is how progress toward reaching goals of understanding is made. Will
humans understand their color vision system in 50 or 200 years, or never? Time
will tell.
The curious thing (or perhaps not) is the considerable practical level of
success that color science has gained in coloration technology, color device
technology and color management with the kind of house of cards models just
mentioned. None of these is perfect and perhaps never will be. One good
reason is that there is just too much variation in individual experience from a
given stimulus set. These relative successes have given color scientists and
technologists a level of confidence that the conjectured relationships might not
be totally coincidental. Coloration of textiles, paints and plastics can now be
controlled at levels approaching just noticeable differences. Metamerism for
a limited number of light sources can be avoided and it is even possible to
obtain a good idea if a matched material sample will or will not have a similar
appearance under different lights.
All existing color spaces and solids result in color experiences that are at
least in ordinal order of perceived color. (It is not known if this applies uni-
versally to color normal observers.) To produce an atlas that is an accurate
representation of a perceptually uniform color solid for a given set of condi-
tions is impossible at the level of large differences. But this does not seem to
be important as long as it is understood in what way the atlas configuration
deviates from the space implied by its samples (as we know we cannot produce
accurate large scale flat geographical maps). Such an atlas has not yet been
produced. The best color difference formulas for small color differences have
limited accuracy for reasons that have not yet been investigated in detail. The
models are likely too simplistic but, as discussed before, there are also many
issues of observer panels and observation conditions. The technical value of
improved formulas makes their pursuit worthwhile.
 WHAT CAUSES THE PERCEPTION OF COLORS AND THEIR DIFFERENCES? 339

9.2 WHAT CAUSES THE PERCEPTION OF COLORS AND

THEIR DIFFERENCES?

Colors are psychological experiences. To explain them reductively with neuro-

logical processes (if that is possible) requires linking propositions that attempt
to create a credible link between subjective experiences and objectively mea-
sured data related to neural responses or physical measures of reflectance (or
spectral power). Today the chain of such arguments built to explain perception
of colors is far from complete. But there is unanimous support for cones as the
sensing devices for light entering the eye at daylight level. The path of the elec-
trochemical signals generated by the cones as a result of light absorptionn leads
through retinal cell layers into the brain and passes through the lateral genicu-
late nuclei as major way stations to visual areas V1 to V4 in the back of the
brain. Area V4 is strongly implicated in color vision, and several kinds of cells
have been identified as responding to color stimuli (up to date details can be
found in Gegenfurtner and Sharpe, 1999, and De Valois, 2000). Here any degree
of certainty ends. V4 signals may be said to disappear into a black box out of
which color experiences appear in an unknown process. S. Zeki and A. Bartels
(1999) believe that conscious color perceptions are generated in area V4 and
may be directly related to (or are) the electrochemical signals in V4. Color cat-
egorization and difference perceptions, as discussed in Chapter 3, appear to
involve additional regions in the front of the brain.
There is growing awareness that, to some extent, our perceptions of forms
and colors are formed through empirical interpretation of the ambiguous
light signals reaching our retinas, in a form of neural networks, based on our
millennia-old experiences as a species and possibly also our experiences as
individuals. A complete vision theory must be able to explain all perceptual
results. It is assumed that by limiting ourselves to simple targets and simple
relativized observation conditions, it might be possible, for these conditions,
to reduce much of the link to between, say, neural events in V4 and perceptions.
In this path and chain of events several types of cells have been identified
that combine in various ways the output of the three cone types over smaller
or larger areas in the retina. Some ganglion cells and cells in the LGN have
opponent character somewhat akin to that suggested by Hering. However, as
seen earlier, there is no direct link between the output of these cells and unique
hues. The implication is that there are additional steps required, for example,
the subtraction of a fraction of the b opponent signal from the a opponent
signal (Chapter 6). In addition the salience of the unique hues may be due to
as yet unknown brain processes, and they may only be indirectly related to a
and b opponent signals. Color cells in V4 do no longer appear to have simple
opponent character, and opponency here may be between cell groups rather
than within cells (De Valois, 2000).
The nature of the relationship between color matching and color appear-
ance is not clear. Color-matching functions change with the size of the test
fields. But there is little change in appearance of most colored materials as a
function of field size. The output of a particular cell type correlates well with
340 CONCLUSIONS AND OUTLOOK

luminance as measured under specific test conditions and is likely to supply

raw data for brightness and lightness perception. As mentioned in Chapter 5,
chromatic perceptions carry with them a brightness component independent
from flicker luminance, giving rise to the Helmholtz-Kohlrausch effect. Bright-
ness discrimination, presumably, is the result of higher-level comparison of the
brightnesses of two contrasting fields, influenced by how our visual system
interprets the scene in which the two fields appear.
Color perceptions fill areas of the visual field surrounded by contrasting con-
tours.A case can be made that color exists only as contrasts, since in the absence
of contrasts, in a so-called ganzfeld view where the eye is covered by translu-
cent, spherical, colored cups, color fades after brief exposure. This implies that
signals from both sides of contours are compared along the visual path and color
experiences are generated as the result of complex computations of the brain
taking into account many aspects of the visual data provided by the eyes.
The fundamental nature of hue and lightness perception is not in doubt.
The degree of contrast of any chromatic perception against a neutral surround,
namely saturation or chroma, appears to be in terms of fundamentality a step
below those primary perceptions, perhaps because achromatic surrounds as
well as saturation or chroma scales are rare in natural scenes.
As discussed in Chapter 6, in simplified conditions hue perceptions are
believed to be formed either as the result of a ratioing process where oppo-
nent signals are compared or as the result of perhaps weighted averaging of
the output of a number of hue detection processes. Hue discrimination, pre-
sumably, is the result of a comparison of two fields with different opponent
signal ratios or different averages of multiple hue processes (see Section 9.11).
While hue perceptions may be the result of a comparison of the ratio of
opponent signals of a field to that of its surround, chroma perceptions may be
the result of the absolute level of opponent signals generated by a field to that
of the surround. Brightness perceptions, in turn, may be the result of a com-
parison of the luminance signal combined with a fraction of the opponent
signals (if any) of the test field compared to the same combination of the sur-
round. All three are affected by specialized processes that modify perceived
hue and enhance or reduce contrast.
By limiting the visual field to a simple set of conditions, the effect of many
of these specialized processes appear to remain relatively constant. In this case
the variables may be limited, for object colors, to the reflectance functions of
the surround and of the two test fields, as long as the display is viewed under
a constant light source and in a standard configuration.

9.3 WHY IS OUR BASIC COLOR EXPERIENCE THREE-

DIMENSIONAL AND WHY ARE THERE FOUR UNIQUE HUES?

These are very fundamental questions in regard to color space. Our immedi-
ate experience of the world as expressed through visual and tactile clues is
 BASIC COLOR EXPERIENCE THREE-DIMENSIONAL AND FOUR UNIQUE HUES? 341

three-dimensional. As discussed before, the geometry of our visual space is

three-dimensional but with a Riemannian geometry. Further the fact of three
cone types involved in color vision argues, on the surface, for a
three-dimensional expression of color experiences. But, as discussed in Chap-
ters 4 and 5, our psychological color space is not directly related to the cone
sensitivity space. There are important transformations, some of which have
only been guessed at so far. In relativized standard conditions the sum of L
and M cone signals is related to the flicker brightness experience. Perceived
heterochromatic brightness has a more complex basis. Hue experiences
may be related to opponent color signals and are correlated with them in ordi-
nal and even roughly interval order. Saturation scales result from additive
mixture of stimuli, causing complementary chromatic, as well as mixtures,
causing chromatic and achromatic experiences. The three cone types are
not the immediate cause of our elementary three attribute color experience.
The cause may be complex cone signal derived mechanisms one of which
appears to result in hue, another in saturation and to some extent brightness
experience, while a third contributes the major portion of the brightness
experience.
In 1964 Judd and co-workers showed that a large variety of measurements
of terrestrial daylight can be expressed with three basis functions. (However,
such reconstruction is only reasonably accurate for correlated color tempera-
tures down to 5000 K, and the corresponding CIE procedure is only applica-
ble to that limit. This limits the applicability to “white” light and excludes the
more strongly colored lights of sunrises and sunsets or colored light reflected
from vegetation, etc.) Shepard (1992, 1994) believes the three basis functions
of natural daylight to be responsible for the existence of our three cone types
and, thereby, for our three-dimensional psychological color experience. This
conjecture appears to be anthropocentric and thus not to explain the reduced
number of cone types in many other vertebrates as well as the spectrally
shifted cone sensitivities of some vertebrate and invertebrate animals. In
the animal kingdom numbers of color-related cone types with different
spectral sensitivities in a species range up to ten (for mantis shrimp; Cronin
and Marshall, 1989).
An alternative conjecture bases the number of cone types in a given genus,
their spectral sensitivities, and the resulting refinement of the information in
the brightness and opponent color systems on specific pressures in their eco-
logical niches and the accidents of random gene mutations. The energy of sun
radiation peaks in the visible region, a good reason for placing receptor sen-
sitivities between approximately 320 to 760 nm. While cone sensitivities found
in nature are limited to this range, the number in a genus may be an empiri-
cal compromise between visual acuity (sharpness of vision) and need (at one
time or another) for spectral discrimination.
The question why four basic hues (and not more or less) seems not yet sat-
isfactorily answered. Hering has not posed it but merely stated the fact that
there are four basic hues and black and white. This number appears also to be
342 CONCLUSIONS AND OUTLOOK

somewhat accidental. If basic hues are a result of opponent signals, then three
cone types, in theory, make six basic hues possible. But because of the high
degree of spectral overlap of the M and L cones, there would be a consider-
able amount of redundancy in the information. Four basic hues were sufficient
to support the survival of our genus.
In the matter of the number of basic hues an intriguing conjecture was
recently offered by Lotto and Purves (2002). They claim that we have four
basic hues in response to the need “to solve the four-color [topological] map
problem” as well as “to order spectra according to their physical similarities
and differences.”
It is evident that broad band sensors can only provide effective discrimi-
nation with overlap of the sensitive regions. Thus, having a short wavelength
sensitive detector added with some overlap to a midwavelength sensitive
detector improved discrimination capabilities in the visible region. Dichromats
can discriminate more than monochromats. Adding a third, overlapping sensor
improved discrimination further. Information manipulation was now required
to avoid considerable redundancy of information and to normalize percep-
tions at different levels of lightness. Based on three cone types the minimal
opponent system has four basic hues. It can be considered evolutionarily suf-
ficient (since that is what we have), but it is certainly not optimal. There is a
large spectral range (from about 580 to 730 nm, nearly 45% of the total visual
spectral range) where there is comparatively little discrimination. On the short
wavelength end the discrimination ability has been significantly improved
(perhaps by up to 20%) by the reappearance of redness. Mathematical analy-
sis shows that filtering of spectral data through the color-matching functions
is not optimal in terms of extraction of the information from the spectral data
and ordering of spectra according to their similarities. Three principal com-
ponents and other techniques do a considerably better job. But cone absorp-
tion is a process with biological limitations, and our visual system represents
a compromise between what is desirable and what is biologically possible. The
question is which evolutionary pressure was larger, that of improving dis-
crimination capability while maintaining a high level of visual acuity or the
unambiguous solution of the four color map problem?

9.4 WHAT ARE THE FUNDAMENTAL PERCEPTUAL

COLOR ATTRIBUTES?

The conventional, but not uncontroversial, view is that they relate to lightness
and hue perception, as well as to intensity of coloration. In terms of evolu-
tionary development brightness/lightness is the oldest perceptual variable.
It antedates the development of color vision. Its current implementation in
humans appears to be of complex nature, fine-tuned to allow useful interpre-
tation of the visual field in terms of light, shadow, and forms. But today it is
not the most salient attribute. This position belongs to hue. Color is defined
 HOW ARE HUE, CHROMA AND LIGHTNESS PERCEPTIONS COMBINED? 343

primarily by hue because of the dramatic nature of our hue experiences.

Natural hue experiences can be subdued in a late fall Northern landscape but
are riotous in a tropical forest. Artificial coloration has expanded the possi-
bilities of strong and varied hue experiences. The importance of hue is further
supported by the fact that a seemingly elaborate (but in detail unknown) appa-
ratus has been put in place to make possible our ability to make fine hue dis-
tinctions. Hue can be described psychologically in terms of the four Hering
unique hues, as Hering, the developers of the Swedish NCS system, Indow, and
others have shown.
Hering was clearly aware of brightness/lightness but did not accord it status
of a fundamental color attribute. Instead, he used blackness and whiteness as
modulators of hue, with a derived attribute of chromaticness. Such a system
has steps, in terms of uniform perceived differences, of considerably varying
size in all directions. Multidimensional analysis of the NCS system using mag-
nitude of perceived difference between steps would undoubtedly arrange it in
a Munsell type configuration with the samples placed at widely varying dis-
tances. Hering type systems that disregard lightness as an attribute can fit con-
ceptually into a simple geometrical solid, as they are not concerned with
uniform differences.
Judging the size of perceptual color differences between objects against a
standard surround—a conventional psychological task—leads to a system
based on hue, lightness, and chroma attributes. If hue and lightness are given
fundamental status for colors of objects in relativized conditions, the third per-
ceptual attribute is chroma, as Munsell discovered. A systematic arrangement
based on uniformity of difference in such a system does not fit into a simple
geometrical solid, nor can it be modeled in a euclidean space, as additionally
discussed below. It apparently has never been tested competitively under con-
trolled conditions which of the attributes, blackness/whiteness or chroma/light-
ness, can be experimentally measured with less variability.
While both types of system are uniform in some sense, it is proposed to
reserve the term “uniform” for uniformity of perceptual distance only and to
use the general term “regular” for others.

9.5 HOW ARE HUE, CHROMA AND LIGHTNESS

PERCEPTIONS COMBINED?

In Nickerson’s Index of Fading the three components, properly weighted,

are simply added. Indow has come to a comparable result based on his per-
ceptual formula. The OSA-UCS data provide no information in this regard
because the basis data for the set involved mixed hue and chroma differences.
Suprathreshold small differences have only been evaluated as mixed hue,
lightness, and chroma differences, and an untested assumption of their euclid-
ean summation has been made (Chapter 5).
Boynton and Kambe (1980) have raised the possibility of hexagonal chro-
344 CONCLUSIONS AND OUTLOOK

matic contours. The chromatic discrimination contours of Krauskopf and

Gegenfurtner (1992) have been fitted with ellipses and circles, but many may
be equally well described by polygons. Superelliptical contours were calculated
by Teufel and Wehrhahn (2000) for threshold data. Rhombic chromatic thresh-
old contours in cone contrast space have been fitted by Strohmeyer et al. (1998)
and by Giulianini and Eskew (1998). Sankeralli and Mullen (1999) tested hue
increment identification using combined hue and contrast increments and con-
cluded that hue increment identification ignores differences in contrast
(chroma) and that the two types of discrimination may be inherently separate.
There is strong evidence that in a polar coordinate, psychophysically
“uniform” chromatic diagram (uniform hue and uniform chroma scales), the
unit difference contours are elongated in the direction of chroma, indicating
that hue difference discrimination is intrinsically more sensitive in terms of
stimulus increments than contrast discrimination. Explicit evidence is found
in the work of vision scientists as well as in threshold and suprathreshold
object color difference data, including the Munsell system. Elongated unit
contours are implicit in the basis data of OSA-UCS, a fact that has been
suppressed in the final system. Such evidence is the basis of Judd’s term
“superimportance of hue.”
The Nickerson and the Bellamy and Newhall evaluations indicate unit con-
tours to become slightly more elongated when the differences become smaller
(2 : 1 at the level of Munsell chips vs. 2.8 : 1 at JND). On average, for small
suprathreshold chromatic differences the ratio is found to be approximately
2.2 : 1 (Chapter 8). The shape is generally taken to be elliptical. Conceptually
it is unlikely to fit the simple geometrical form of an ellipse.
How chromatic and lightness differences are combined is also a not a ques-
tion that is resolved. Recent neurological investigations show that “interac-
tions between color and luminance variations can be complex and not easily
predictable” (De Valois, 2000). In terms of color difference formulas the
general assumption has been that the sum is euclidean. Based on his color-
matching error work MacAdam maintained that the resulting ellipsoids
have horizontal alignment (e.g., MacAdam, 1981; Section 8.4). Threshold and
suprathreshold color differences are fitted better with formulas that result in
an ellipsoid tilted at higher lightness in the direction of the neutral axis, as is
the case with CIELAB and related formulas.

9.6 WHAT CAUSES THE PERCEPTION OF THE MAGNITUDE OF

COLOR DIFFERENCES?

At this time the answer to this question, in neurophysiological terms, is

not known. From findings in Chapter 8 we can conclude that at the color-
matching error level differences are largely due to sensitivity limits of the cones.
This is fully the case for the MacAdam data and largely for the Wyszecki-
Fielder data. At the threshold level, based on the Richter data, an opponent
 CHRISPENING EFFECTS 345

color system seems fully engaged, as evidenced by the lower S/L, respectively
Z/X, ratios. This is the case for larger color differences up to and, likely, ex-
ceeding the size of OSA-UCS differences. Here power relationships connect
stimulus and perceived difference. The applicable powers and the weighting
constants (different by semi axis) are found to vary widely in different exper-
iments for reasons that are not clear but may involve the design of the exper-
iments or the observer panels. Different optimal powers and weights appear
to indicate independence of at least four chromatic processes. The immediate
reasons for the power relationship are not known but assumed to be satura-
tion effects of various cell types.
The relative enlargement of unit L increments against unit S increments for
threshold and larger differences compared to CME data signals a sea change,
the change from the Helmholtzian cone level to the higher zone, opponency
level. The higher intrinsic sensitivity of the L and M cones compared to S is
suppressed. Regardless of the input from cones, here one unit of a is equal to
one unit of b, with the positive and negative functions in balance. The relative
number of L, M, and S cones per unit area in the retina does not seem to play
any role in this situation. The new ratio of implicit L versus S response of the
first step from gray is valid throughout the range of differences from thresh-
old to OSA-UCS sized (and very likely larger) differences.

9.7 CHRISPENING EFFECTS

Crispening effects impose significant modulation on the relationship between

stimulus and perceived colors or color differences. They are the result of an
environmentally useful adaptation that makes minor spectral differences of
objects against surround especially well discernible. As discussed in Chapter
4, stimulus increments/decrements resulting in a criterion difference percep-
tion are smallest if the test field stimuli straddle that of the surround color.
The more the surround differs in hue, chroma, or lightness from the test colors,
the larger will be the required stimulus increment for a criterion difference
response between the test colors. Highest chromatic discriminability is thus
available in a chromatically relatively homogeneous environment, and com-
parably for lightness discrimination. Like lightness crispening, chromatic
crispening may depend on test field size. Detailed quantitative data are not
available in both cases.
Crispening effects result in V-shaped functions of unit increment versus
stimulus (L, M, S or X, Y, Z) with the vertex pointing at the surround point
(Chapter 8). In these functions different observers and, presumably, different
observation conditions result in different angles between the legs of the V
function. Functions for brightness and lightness data and for the S (or Z) cone
have larger angles than functions for the L and M cones.
Lightness crispening in simple surround conditions is controlled by the
luminous reflectance of the surround and, probably to a lesser extent, by the
346 CONCLUSIONS AND OUTLOOK

test field size. Increments for a criterion response become increasingly larger
as the test field luminous reflectances differ from that of the surround. Color
space and difference formulas so far have generally not considered this
effect. If achromatic reference pairs are used the magnitude of difference per-
ceived between them is also subject to the lightness crispening effect, that is,
the psychological magnitude of the difference between two gray reference
samples depends on the surround and the relative size of the fields. This issue
affects comparability of different data sets based on different surrounds.
The lightness crispening effect is active from threshold size differences to
differences of the size found in the Munsell value scale and, likely, larger
differences.
The chromatic crispening effect is, on a relative basis, about equally distinct
for L and M compared to S. The L increment required for criterion response
quadruples approximately over a change in L of 10 units from the neutral
point in direction of increasing L, less so for decreasing L, as the increment is
a sum of the fundamental increment and the chromatic crispening increment.
As has been shown in Chapter 8, it is most strongly active at the level of CME
and threshold differences and appears to gradually fade as chromatic differ-
ences become larger. It is absent at the level of Munsell double-chroma steps
or OSA-UCS steps.
Chromatic crispening makes it possible to euclidize the Riemannian space
implicit in small color difference data. The approximate factor 2 of the ratio
of the longer (chroma-related) to the shorter (hue-related) diameter of the
unit contour has been included in the SC weight adjusting for chromatic
crispening. Since this weight is a continuous function of chroma it can be inte-
grated in the euclidization of formulas like CIE94.

9.8 PERCEPTUAL INCREMENT MAGNITUDE AS A FUNCTION OF

STIMULUS INCREMENT MAGNITUDE

Despite the findings of Plateau the question of increment magnitude for a cri-
terion response as a function of stimulus magnitude was dominated for the
second half of the nineteenth century by Fechner and the Weber-Fechner law
indicating the required increment to be a constant fraction of the stimulus.
Study of lightness scales from black to white in the early twentieth century
and additional investigations of other sensory magnitudes pointed to the
applicability of power modulation. Different experimental conditions resulted
in different optimal power modulations. The findings in regard to crispening
discussed in Chapter 8 indicate that the Weber-Fechner law is generally not
applicable to color differences. It is also apparent that simple power functions,
as proposed by Stevens, are not representative of color differences from sub-
threshold to medium size.
In global psychophysical color space scaling efforts the Munsell value scale,
 HOW WELL DO FORMULAS PREDICT PERCEIVED COLOR DIFFERENCES? 347

with its suppression of the lightness crispening effect (and its fit with cube root
modulation) and the proposal by Adams in 1942 to apply cube root modula-
tion also to the X and Z tristimulus values, proved influential and continues
to be the official paradigm of the CIE. This situation has been influenced in
the past by computational difficulties and then by habit. The evidence derived
from the Munsell Renotations and the Re-renotations is that different powers
are optimal for the four chromatic semi axes. Conceptually it would be sur-
prising if the complex processes of the visual and related cognitive systems
could be accurately modeled with a single and simple power modulation. This
matter has not yet been thoroughly investigated in regard to threshold and
suprathreshold small color difference data.
Experimentally determined incremental and decremental stimulus
amounts lack a solid foundation in neurophysiology. The prevailing explana-
tion involves cone response saturation effects. It is unlikely to be the complete
story. The situation is complicated by the fact that psychophysical models need
to account for three effects in the chromatic plane:

1. Basic nonlinearity of the tristimulus space. Unit contours in the nor-

malized X, Z diagram are elongated. They are largest and most
elongated when Z values are large. The contours are smallest for
yellowish-greenish colors. Power modulation of the scales linearizes
them and makes the ovoid contours of small color difference data at a
given chroma approximately equal in size. It makes the ovoid contours
of Munsell data approximately equal in size throughout the diagram.
2. The chromatic crispening effect changes the size of contours of small
color difference data as a function of chroma. The power-modulated
(i.e., normalized in the four quadrants) contours differ in size by a factor
of 4.4 : 1 as a function of chroma in a metric chroma range from 0 to
100 (Chapter 8).
3. The unit contour in a euclidean diagram is fundamentally elongated and
the elongation may be a function of the size of the differences, as seen
in the previous section.

In formulas such as CIE94 and CIEDE2000 the cube root formula addresses
point 1. The weight on chroma differences and the chroma-related weight on
hue differences addresses points 2 and 3.

9.9 HOW WELL DO FORMULAS PREDICT PERCEIVED

COLOR DIFFERENCES?

As seen in Chapter 6, correlation between average visual data and formulas

can be improved to some extent with empirical modification based on regu-
larities observed in the data. Such modification is often specific to a given data
348 CONCLUSIONS AND OUTLOOK

set. When looking at agglomerations of data approximately two-thirds of the

variation in average perceived difference is predicted by the best formulas.
The performance factors PF/3 (a performance measure approximately equiv-
alent to error percentage) of the combined data set used in the development
of CIEDE2000 as well as of various subsets assembled from different original
data sets (hue, chroma, and lightness differences, as well as others) are uni-
formly from 30 to 34. Similar values have been obtained with that formula for
specific data sets such as BFD, Witt, and BIT. The only exception is the RIT-
DuPont data with a value of 19. While the PF/3 results for the BFD data
improve from 56 for CIELAB to 37 for CIEDE2000, the improvement for
RIT-DuPont is only from 22 to 19.
Progress in the accuracy of prediction can only be made if the source(s) of
the error component representing one-third of the total variation of the
formula can be determined. In an analysis of five sets of data for several of
them significant increases in correlation were obtained by adjusting the size
of the perceived difference by size factors seemingly nonsystematic for each
color subset (Kuehni, 2001b). No agreement in these factors for given color
regions in the different sets could be found. The factors ranged from approx-
imately 0.7 to 1.5. This indicates that the size of average judged difference in
a given color region varies significantly in different experiments despite efforts
at normalisation. At this point the changes appear to be random, but detailed
investigation into the potential causes has not yet been made. When the
remaining major discrepancies in the RIT-DuPont set are analyzed, they
appear to involve primarily complex differences with hue, chroma, and light-
ness components, hinting at non-euclidean addition. Among other potential
causes are observation conditions such as lighting and surround, composition
of observer pool, subjective strategies of the observers, method of averaging
individual results, choice of reference difference, and nature of dividing line
between samples.

9.10 IS UNIFORM COLOR SPACE EUCLIDEAN?

Psychophysical models provide fits of greater or lesser degree to psychologi-

cally uniform color difference data. But euclidean depictions of psychological
color space are not uniform, unless (for small color differences) achieved by a
special euclidization step. The Munsell chromatic diagram is approximately
uniform in terms of hue (at one chroma level) and in terms of chroma, but not
both. If the scales are adjusted to equality the chromatic plane can only be illus-
trated as a crinkled fan. It is not known how far reaching Nickerson’s investi-
gation of the relationship of hue and chroma differences were as a function of
chroma. If the SH weight is a truer representation of facts than an assumption
of a constant ratio between hue and chroma differences as in the Nickerson
formula, the uniform color space is Riemannian. As discussed, a portion of SC
 IS UNIFORM COLOR SPACE EUCLIDEAN? 349

(together with SH) eliminates elongation of the unit contour. It is not clear that
it actually does so because the complete hue circle has not been scaled rela-
tive to chroma in connection with color difference formulas that have such
weights.
Euclidean structure has been denied by Judd (1968) in case of a space based
on a constant achromatic surround with arguments involving the superimpor-
tance of hue and crispening effects. An additional argument was called “dimin-
ishing returns in color difference perception.” MacAdam had pointed out in
1963 that if there are three color samples A, B, and C selected along a geo-
desic so that samples A and C appear equally distant from sample B, samples
A and C are seen as having less than twice the difference between A and B
or B and C. In other words, the applicable scale is an interval scale but not a
ratio scale. This effect may be connected with the fading of chromatic crispen-
ing for larger differences and the changing of the unit contour elongation ratio
as a function of size of the difference.
In 1995 Wuerger and co-workers reported on tests of euclidean color geom-
etry. In one test they used proximity judgments to determine the angle
between intersecting lines in color space, and three kinds of tests for the addi-
tivity of such angles were made. All three failed to support a euclidean struc-
ture. In a second experiment the increase in variability of judgments of
similarity with the distance between test and reference stimulus was deter-
mined. Also this experiment failed to support euclidean structure.
The euclidean addition of hue, chroma, and lightness differences does so far
not seem to have been tested explicitly at the level of suprathreshold small or
large differences and under conditions representative of industrial practice (no
control of local adaptation). Among the reasons are the difficulty in preparing
accurate, well defined samples and the lack of replicated uniform chroma, hue,
and lightness (including the Helmholtz-Kohlrausch effect) scales.
As mentioned, euclidean structure with luminous reflectance as one dimen-
sion also can be expected to fail because visual lightness perception of chro-
matic samples involves not only luminous reflectance but also the presumed
contribution of the opponent color system in the Helmholtz-Kohlrausch
effect. Resulting constant psychological lightness planes are not orthogonal to
the achromatic lightness axis.
The question arises if it is useful and veridical to appropriately transform,
for small color differences, a* and b* scales by integration so that the result-
ing diagram is of euclidean nature, as Thomsen and others have shown and as
implemented in the DIN99 formula. Given the many remaining problems,
some addressed in CIEDE2000, the result is likely less than satisfactory and
not useful for large color differences. Another version of integration used in
the past is projective transformation, extensively employed by MacAdam.
However, his best effort in devising a nonlinear transformation for the OSA-
UCS basis data did not result in correlation improved over that obtained with
a cube root based euclidean formula.
350 CONCLUSIONS AND OUTLOOK

9.11 UNIQUE HUES AND UNIFORM COLOR SPACE

A psychological euclidean color space with the unique hues on the axes is not
uniform in terms of differences because hue differences representing identi-
cal hue angle differences are of different perceptual magnitude in the four
quadrants (Chapter 4). In addition there is the matter of the elongated unit
contours.
There is poor agreement between the cone contrast chromatic diagram axes
and unique hues. There is only slightly better agreement in case of a chromatic
diagram based on opponent functions derived from CIE color matching func-
tions. But there are two major issues here also:

1. The implied unique hues of the two standard observers, taken as the
crossover points of the opponent functions, are considerably different.
This implies that the primaries represented by the system axes are of
more or less mixed hues for both standard observers.
2. In terms of both standard observer opponent color functions objects
seen on average as unique red and unique green have a considerable
positive b (“yellowish”) component.

No form of transformed color matching and related opponent functions seems

to exist where the four average perceived unique hues fall on the axes and a
constant perceived chroma circle is represented as a circle (Chapter 5).
The processes behind unique hues are unknown and may have a location
different than opponent processing. But even when known, there remains a
systematic discrepancy between a space based on unique hues and their stan-
dard increments and decrements and a uniform space.

9.12 EVIDENCE FOR THE OPERATION OF AN OPPONENT

COLOR SYSTEM

Aside from the neurophysiological evidence and the purely psychological evi-
dence in the Hering sense, there is considerable indirect psychophysical
evidence. In a cone-sensitivity space an approximately perceptually uniform
system such as the Munsell system is placed in alignment not in agreement with
any of its axes (see Fig. 5-29). Rough alignment with the axes is only obtained
in a tristimulus space and constant chroma circles at different lightness levels
only become concentric if the chromatic diagram is based on opponent func-
tions (Fig. 5-31). In other words, a reasonably accurate model that places the
information contained in reflectance functions of the Munsell system into a
corresponding psychophysical model requires opponent color functions.
Implicit evidence for an opponent system also comes from the fact that cone
increments required for the perception of threshold and suprathreshold
 OPPONENT SIGNALS: THE SOURCE OF HUE AND CHROMA PERCEPTIONS? 351

differences are, in case of L and M cones, not related to their intrinsic sen-
sitivity as expressed in color-matching error but represent increments of
approximately double relative magnitude. Such enlarged increments are
consistent with the operation of an opponent system.
Perceived brightness/lightness of chromatic colors is not in agreement with
power-modulated flicker luminance/luminous reflectance. But it can be
modeled closely with addition of a portion of opponent color activation to
flicker luminance (Chapter 5), indicating that perceived brightness/lightness
of chromatic colors is not a dimension of tristimulus color-matching space.

9.13 OPPONENT SIGNALS: THE SOURCE OF HUE AND

CHROMA PERCEPTIONS?

In Chapter 7 evidence was offered that it is not possible to represent the

Munsell system as a uniform euclidean color solid because of the superim-
portance of hue. Because of chromatic crispening the effect of hue superim-
portance can be integrated in small color difference data to end up in a
euclidean system.As mentioned, this may not be possible at the size of Munsell
color differences where chromatic crispening no longer applies. This fact may
be a consequence of our assumption that output of hue and chroma discrim-
ination processes is caused by the same opponent color signals. Such mecha-
nisms, while believed reducible to increments of opponent color responses,
represent separate entities. Hue and chroma perception occupy the same two
dimensions in a chromatic diagram. They would have to be balanced in terms
of the relative perceptual magnitude of the unit stimulus increments to fit into
a euclidean system. But as seen in Chapter 7, they are not.
Can opponent color signals be the immediate source of hue and chroma
perceptions? In other words, is the hue system based on opponent signals
only? As seen in Chapter 4, when implied opponent signals of an ideal hue
circle are plotted as a function of hue angle they form offset sinusoidal curves.
Hues that represent unit hue differences from selected reference colors along
the constant chroma circle plot on a line parallel to the sinusoidal line of the
circle (Fig. 9-1). For differences to be hue differences only, the following rule
applies: the sum of a and b of the test must be identical to that of the refer-
ence; that is to say, differences in a and b between reference and test must be
opposite in direction and identical in magnitude. Colors that do not follow this
rule result in either chroma or mixed hue and chroma differences. For a pure
chroma difference, differences in a and b between reference and test must not
only be of the same magnitude but also the same direction. Mixed hue and
chroma unit differences of a given magnitude are defined by a and b values
that are bounded by the pure hue and pure chroma differences. In such a
scheme it is difficult to see where the elongation factor 2 of unit perceived dif-
ference contours originates. It seems likely that different mechanisms are
responsible for hue and chroma perception (certainly also lightness percep-
352 CONCLUSIONS AND OUTLOOK

Fig. 9-1 Values of an ideal, euclidean, constant chroma first quadrant with identical hue
differences between adjacent colors. Circles represent the reference color, and triangles the
test color.

tion) and that additional processing takes place beyond simple opponency. The
real hue and chroma perception and hue and chroma difference perception
processes may be more complex as hinted at by the results of Chichilnisky and
Wandell (Chapter 6).

9.14 THE APPROXIMATE SHAPE OF A UNIFORM COLOR SOLID

Given the difficulties alluded to in earlier sections and the lack of sufficiently
detailed experimental data it is not possible to create an accurate geometrical
model of a uniform object color solid. An approximation of the surface of such
a space based on small color differences was calculated using simplifications
and assumptions. At eight different levels of luminous reflectance twenty
maximal object colors around the hue circle were determined. The luminous
reflectances were adjusted to reflect the Helmholtz-Kohlrausch effect using
the corresponding formula from OSA-UCS. These values were transformed
into the L*, a*, b* space and the a* and b* values euclidized using the Thomsen
formula. As was mentioned in Section 9.9, it is not clear if this calculation
adjusts also fully for the fundamental elongation of the unit contour. The
results are points on the surface of a color solid uniform according to the con-
ditions implied and are shown in three projections in Fig. 9-2a, b, and c. View
9-2a shows the solid facing the yellow to blue dimension and view 9-2b facing
 A RESEARCH PROGRAM 353

Fig. 9-2 View of a color solid approximately uniform at the level of small color differences.
The vertical line connects achromatic colors from luminous reflectance 0 to 100. The different
symbols represent colors of a given luminous reflectance at the object color limit. They have
been adjusted for the Helmholtz-Kohlrausch effect. (a) View of the yellow-to-blue axis; ( b) view
of the green-to-red axis; (c) view from the top onto the chromatic plane.

the green to red dimension. Achromatic colors fall on the straight vertical line.
The third view shows the solid from the top. Its irregular nature is clearly
evident. It can be compared to Fig. 1-2.

9.15 A RESEARCH PROGRAM

Analysis of published data in regard to color attribute scaling and color dif-
ference scaling shows significant discrepancies in results between different
efforts, despite large numbers of samples, observers, and observations. There
are several potential reasons for such discrepancies: different observer panels,
different visual context of samples and surround, different light sources, and
354 CONCLUSIONS AND OUTLOOK

Fig. 9-2 (Continued)

unrecognized contextual clues that can affect the results due to empirical inter-
pretation of the visual field.
Independent replication of experimental results is a standard requirement
in science. The degree to which experimental color scaling results can be inde-
pendently replicated needs to be established. Given the known and perhaps
some unknown sensitivities of results to experimental conditions, it seems
imperative that independent replication involve essentially identical viewing
conditions: the same samples, the same visual context, and the same or very
similar (implicit or explicit) light source. Today duplication of this kind is prob-
ably best achieved with simulated samples on a computer monitor. The same
program creating the displays can be used in different locations, and the
general surrounds can be duplicated with masks surrounding the monitor
screen. Illumination of the room in which the monitor is viewed may be impor-
tant for control of the complete adaptation situation. Monitor calibration is a
critical issue that needs to be satisfactorily resolved. Variability of results from
different observer panels in several locations can then be established. The
results may indicate if the variability we see in past results is normal or if, given
 A RESEARCH PROGRAM 355

Fig. 9-2 (Continued)

tightly controlled test conditions, the variability can be substantially reduced

and how many observers are needed to obtain a reliable average. Alternately,
identical physical samples could be used in several locations with closely dupli-
cated observation conditions but different observer panels. Global scaling with
Munsell sized differences should be followed by scaling of smaller differences.
Given the clear indications of hue superimportance, it appears that global
scaling should be based on lightness, chroma, and hue scaling. Consideration
has to be paid to the question of the reference difference used to ensure that
lightness, chroma, and hue steps are of identical perceptual magnitude. In
closed scales such as lightness and hue, selection of a particular size of refer-
ence difference may result in a number of steps that is not an integer.

Lightness Scaling
Once an intermediate lightness for the achromatic surround and a reference
difference have been chosen, a 10- or 20-step gray scale can be established.
356 CONCLUSIONS AND OUTLOOK

Planes of perceived iso-lightness in a linear a, b, Y space can then be deter-

mined at different levels of Y of the gray reference. This is not unproblematic
because of the findings of Nayatani and co-workers (1994) that the magnitude
of the Helmholtz-Kohlrausch effect depends on the experimental design. The
reason for the significant differences reported by these researchers should be
elucidated and the experimental design selected accordingly. Iso-lightness
planes need to be determined at different levels of lightness. It is unlikely that
the assumption of Wyszecki and Sanders and the OSA-UCS committee that
planes of iso-lightness as a function of Y are parallel over a wider range of Y
is valid.

Chroma Scaling
Once planes of constant lightness have been established chroma can be scaled
at different levels of chroma and different levels of perceived lightness. Again,
the work may be easiest to accomplish using monitor display colors that are
adjustable essentially in one attribute direction only. As the observer adjusts
fields for identical perceived chroma, the program should automatically make
adjustments for the previously established constant lightness and follow
approximate lines of constant hue as reported in the literature. Once 20 to 40
hues around the hue circle have been scaled for chroma, constant chroma
contours can be established by interpolation of data points. Such contours
need to be established minimally at 5 to 7 levels of chroma at several levels
of perceived lightness.

Hue Scaling
With chroma scaling complete, constant chroma contours can be scaled for
constant hue increments along the contours. Fine adjustments can be made to
result in an integer number of differences. Again, the program should make
automatic adjustments for perceived lightness and chroma as the observer
makes changes of the perceived hue of the test field.
Many issues need to be decided. For example, should all three attributes be
scaled separately for each observer, or should average perceived lightness and
then average perceived chroma be used for the subsequent step?
Once results of the best achievable level of replication have been obtained,
efforts to fit the visual data with psychophysical formulas can be undertaken.
It remains to be seen if formula fitting achieves satisfactory agreement or if,
in the end, lookup tables are required.

Complex Color Difference Scaling

With global scaling at the Munsell level complete efforts can be undertaken
to determine the addition process required for complex color differences: Is
 KINDS OF SPACES 357

the addition of hue, chroma, and lightness differences simple, euclidean, or

some other geometry? This issue has to be studied for pairwise complex dif-
ferences as well as for differences ranging through all three attributes. The
replicated findings at the Munsell level of difference need to be confirmed or
modified for small suprathreshold differences.
The apparent increase in chromatic crispening as the size of difference
decreases requires quantification, both in regard to hue and chroma. Clarifi-
cation is also required of the apparent nonlinear effect of lightness on the size
of perceived chromatic differences, as discussed in Section 8.6. Additional
quantification of the lightness crispening effect in small color difference eval-
uation is also desirable.
The form of the functions relating attributes and size of differences and the
answer to the question of type of addition will shape the form of an equation
for small suprathreshold color differences. Also here it remains to be seen if
an analytical formula is useful or if the information should be used in form of
lookup tables. The same question applies to a universal color difference
formula. Can such a formula be meaningfully expressed with equations with
changing parameters?

Robustness of Formula
The robustness of a formula developed in such a manner needs to be estab-
lished. We have already seen that without several adaptable parameters such
a formula is only strictly applicable to a narrow set of conditions in regard to
samples, surround, and lighting. The question to be resolved is how much con-
ditions can change before the accuracy of the formula is unacceptably reduced.
This may involve sample size and structure, separating line, surround quality
and structure, spectral power distribution, and intensity of the light source. Can
the formula be meaningfully adapted to apply for incandescent or triband
lamps?

9.16 KINDS OF SPACES

This text has made clear that the idea of a single fundamental color space is mis-
placed. As found in practice there are different kinds of color spaces and solids
applicable with a degree of accuracy to different situations. Changes in the
applicable conditions can quickly further reduce the accuracy of a formula to a
significant degree.
Color spaces can be placed in a kind of hierarchy of complexity and
meaning. The structural requirements for these kinds of spaces differ.

Uniform Color Appearance Spaces and Solids

These are spaces that for a specific relativized set of viewing conditions illus-
trate object color samples not only embodying qualitative systematicity in
358 CONCLUSIONS AND OUTLOOK

arrangement of colors according to three fundamental attributes but also

quantitative systematicity according to the principle of visual uniformity of
difference in all directions. Such a space, as we have seen, is non-euclidean in
nature but (at least for small color differences) can be euclideanized. For
spaces of this kind a high degree of correlation between psychological and psy-
chophysical data is important. Their primary use is for quality control pur-
poses. There is currently no sample set that can be said to satisfactorily express
a uniform color solid for the average color normal observer and the applica-
ble set of experimental conditions. Thus all currently existing physical sample
sets must be classified as regular, rather than uniform. As a result there is no
psychophysical space or color difference formula with satisfactory perfor-
mance either, as seen earlier. It would be of considerable interest to know how
high a degree of correlation can be obtained for an individual observer, given
knowledge of that observer’s personal cone sensitivities, luminance function,
and other properties of the visual system. It might then become apparent to
what degree individual experiences shaped our personal color vision appara-
tus in a way that cannot be systematically expressed with the kinds of sim-
plistic color vision models currently at our disposal. Should the effect be of
significant magnitude a color difference formula with much improved perfor-
mance for the average observer may be impossible. Our current level of
knowledge is insufficient to come to such a conclusion.

Regular Color Appearance Spaces

Much of what has been said in the previous subsection is applicable here also.
In case of the NCS system differences in principal hue components, blackness
and chromaticness are not linearly related to perceived magnitude of differ-
ence. Close agreement between the NCS psychological space and a psy-
chophysical space is not important, and in fact no complete mathematical
formulation has been attempted so far. Whenever translations seem necessary,
they are made with the help of tables. Regular as well as uniform color solids
are equally reflective of and variable with the experimental conditions. OSA-
UCS, Munsell, and DIN are roughly uniform but euclidean. Ostwald and NCS
have also quantitative systematicity but according to principles other than uni-
formity. Colorcurve, HSB, RGB, CYMK, and other systems have value as
arrangements with ordinal systematicity with different kinds of stimulus incre-
ments: tristimulus increments in case of Colorcurve, electron gun output incre-
ments in case of HSB and RGB as related to video display, primary print color
increments in case of CMYK. In the limited region of a color solid that can
be covered with colorants or monitor phosphors, they are indicative of the
variety of color experiences of the average color normal observer under the
limitations of the system. OSA-UCS with its crystalline structure can display
a wide variety of color scales varying simultaneously in all three perceptual
attributes and changing in color in dramatic ways without, as shown, being
uniform.
 KINDS OF SPACES 359

Psychophysical (Regular) Spaces

For spaces such as Luther-Nyberg, CIELUV, CIELAB, and others there are
smaller or greater claims for uniformity, but they are just regular spaces
with different principles of tiling. A particular example is the Derrington-
Krauskopf-Lennie space with tiling according to macaque cell response in the
lateral geniculate nucleus.

Color-Matching Spaces
Color-matching spaces include many examples, such as the Rösch-MacAdam
space, the CIE tristimulus space, and the Cohen fundamental color space (with
many possible configurations). The last derives its name from the decomposi-
tion of spectral power distributions into the so-called fundamental of each
reflectance function, meaning the portion of each spectral function free of
metameric black. Each related color perception is represented as a vector, just
as in the other two spaces mentioned. Average visual metamers plot in the
same location in all three cases. These spaces make no claims for quantitative
systematicity in terms of appearance.

Spectral Spaces
Spectral spaces do not involve data related to human color vision and, there-
fore, should not be termed color spaces. They are based on finding components
representing a given reduced dimensionality implicit in the spectral functions,
such as by principal component analysis. Some three-dimensional spectral
spaces place spectra of the Munsell colors in correct ordinal order, but the
dimensions have no perceptual meaning and differences between colors in
such spaces have no meaningful relationship to perceived differences. Visual
metamers plot in different locations in such spaces.

Among the many kinds of color spaces the uniform space has the highest
degree of systematicity if equal sense distance has the importance that psy-
chophysics assigns to it. What degree of accuracy and repeatability is possible
for a uniform color space for the average observer and a specific set of exper-
imental conditions remains to be determined. Unless one wants to raise a
charge of bias of some kind for the RIT-DuPont data, they show that signifi-
cant improvement compared to many other data sets may be possible based
on careful experimentation. Additional improvement resulting from a fuller
understanding of the color vision process may also be feasible. Problems and
uncertainties increase the more complete the color vision model is to be in its
explanatory power, for example, as color appearance models indicate.
Given the view that vision is an animal type’s best way to deal with the
ambiguities of the information provided by the energy streams that we call
light, it is evident that finding the one color space representative of human
360 CONCLUSIONS AND OUTLOOK

color vision is a problem without solution. Taxonomy, the classification of

objects and phenomena, is an important human activity. Just as any other
human activity it is open ended. We have found that a color classification free
of context is of necessity very vague. To make it less vague, we must specify
the context but the classification becomes applicable only to that context. To
achieve the highest level of classification (a uniform color space), the context
must be highly defined in all respects.
After some 300 years of effort the conclusion is that color appearance
systems such as the Munsell system are general indicators of the multitude of
color experiences that can be generated, for example, with pigments painted
on paper and viewed in daylight. To reach the borders of our possible experi-
ences work with monochromatic lights is necessary. But it is evident that claims
for a given relationship between stimulus and resulting experience are only
applicable to closely circumscribed conditions of viewing. A uniform color
space (not euclidean) defined in terms of stimuli is only possible for a closely
circumscribed set of viewing conditions and in theory only applicable to one
observer.
 Notes

Chapter 1
1. For empirical rules, see Purves and Lotto (2002). For an elementaristic approach,
see Mausfeld (1998). For relativized conditions, see McLaughlin (2002). On the
change in appearance of objects, consider the works of Impressionist painters
such as those of Monet who distinctly varied the coloration of his Haystack
series and views of the Cathedral of Rouen depending on time of day and/or
weather.
2. On the ordinal order of personal spaces, it is well known that only a small per-
centage of color normal observers perform error-free in the Farnsworth-Munsell
100-hue test where colored caps with small differences in hue must be sorted in
correct sequence.
3. For unique hue variation, see Kuehni (2001a). In the Munsell Book of Colors the
complete hue circle has been divided into 40 perceptually equally different hue
steps.
4. On the four cone types in women, tetrachromacy is well established in the animal
kingdom. The possibility of tetrachromacy in human females was raised by Jordan
and Mollon (1993). Jameson et al. (2001) have investigated the color experiences
of females with the genetic potential for having four cone types. The estimate is
from Neitz et al. (1998).
5. On adaptation, see, for example, Fairchild (1998). Current usage of the term in the
vision science community is limited to processes beginning in the retinal layer. On
retinal illumination, actual levels of illumination at the retina are extremely diffi-
cult to determine. Instead values based on luminance arriving at the eye, assumed
pupil size as well as other assumptions are calculated. The unit of retinal illumi-
nance is the troland.

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.
361
362 NOTES

6. Lux is the photometric unit of illuminance, 1 lux = 1 lumen per m2; lumen is the
unit of luminous flux. A lumen is equal to the flux emitted in a unit-solid angle
from a uniform point source of one (standard) candle.
7. CIE is the French acronym for International Commission on Illumination, an inter-
national body concerned with technical aspects of lighting and color. For more
details on the Munsell color system, see Chapters 2 and 7. The Munsell system is
perhaps the best-known color appearance system.
8. The Nickerson-Newhall psychological color solid models are located at the Hagley
Museum and Library in Wilmington, DE. The models have been manufactured by
Nickerson’s assistant K. F. Stultz.
9. CIELAB is a color space and difference formula recommended by the CIE; see
Chapter 6.
10. On Schönfelder’s law, see Schönfelder 1933.
11. The broad use of the word sensation in the historical psychophysical literature
makes it impossible to use uniform terminology for these two terms.

Chapter 2
1. A review of color appearance spaces was provided by G. Wyszecki in 1960 and
G. Derefeldt in 1991. Information is also provided on the Web site
www.colorsystem.com.
2. Xenophanes as quoted in Freeman (1957).
3. Pythagoras as quoted in Mansfeld (1986), translation by the author.
4. Quote from Empedocles in Mansfeld (1986), translation by the author.
5. Quote from Democritus in Mansfeld (1986), translation by the author.
6. Photius as quoted in Gage (1993).
7. Pliny as quoted by Heinrich Meyer in Goethe, Geschichte der Farbenlehre, 1810,
translation by the author.
8. Ancient Greek sacral colors, Oxford Classical Dictionary, 3d ed. New York: Oxford
University Press 1996.
9. On Avicenna scales, see Gage 1993.
10. On Eraclius, see Merrifield 1967.
11. On Urso de Salerno, see Gage (1993).
12. Albertus Magnus quoted in Gage (1993).
13. For an analysis of Bacon’s work on color, see Parkhurst (1990).
14. For Cennini quote, see Cennini (1933).
15. Leonardo, Codex Urbinas latinus 1270, see McMahon (1966).
16. For Ficino list, see Barasch (1978).
17. For Telesio list, see Goethe (1810), translation by the author.
18. For Cardanus list, see Barasch (1978).
19. Forsius translation and figures from Feller and Stenius (1970) and Parkhurst and
Feller (1982).
20. Boethius and figure, see Murdoch (1984).
 NOTES 363

21. On Oresme, see Murdoch (1984).

22. On Fludd, see Parkhurst and Feller (1982).
23. Newton was not explicit about how he arrived at seven primary colors. It is pos-
sible that he saw seven major hues in the spectrum. It is somewhat surprising that
turquoise or sea blue is not one of his primary colors, given its prominence in the
spectrum. On the other hand, seven was a classical number, and there is the con-
nection to the musical scale that Newton referred to. Perhaps he was consciously
or unconsciously influenced by such associations. However, in a recent paper
Jameson et al. (2001) show that 52 trichromats, when viewing the spectrum with
both eyes, delineate it on average into 7.4 hue ranges.
24. Descartes’s figure in Gage (1993).
25. Diderot quoted from Diderot (1798).
26. On C.B., see Parkhurst and Feller (1982) and Gage (1993).
27. On Castel, see Gage (1993) and Schwarz (1999).
28. On color of urine diagrams, see Gage (1993).
29. For a German translation of Mayer’s paper, see Lang (1980). For an English trans-
lation, see Fiorentini and Lee (2000).
30. The name gamboge is from the word Cambodia, it is a yellow-colored gum resin
from trees of the genus Garcinia, growing in southeast Asia.
31. For more on Frisch, see Schwarz (1999).
32. The English glassmaker George Palmer (1740–1795) had an interest in color and
was active at times in France. In 1777 he published in English a pamphlet titled
Theory of Colours and Vision. It was translated in the same year into French. Walls
(1956) considers Palmer’s theory “just as complete as Young’s, and nowise inferior
to it. . . .” Palmer’s theory was reviewed in Lichtenberg’s Magazin in Göttingen,
Germany, in 1781. Thomas Young was a medical student in Göttingen from 1795–96
and knew Lichtenberg who had considerable interest in vision. It is not known if
Young learned of Palmer via Lichtenberg. In 1786 Palmer also published (in
French) an account of color vision deficiencies.
33. Runge translations by the author.
34. On Matthias Klotz, see Schwarz (1999).
35. On early development of psychophysics, see Boring (1929).
36. Chevreul translation from the English edition of 1854.
37. On Doppler, see Schwarz (1992).
38. Aubert quote from Aubert (1876).
39. See Donders (1881). See also Turner 1994.
40. On further development of psychophysics, see Boring (1929).
41. On Munsell system development, see Munsell (1918) and Kuehni (2002a).

Chapter 3
1. On linking propositions, see, for example, Teller and Pugh (1983).
2. For the nine-dimensional universe, see, for example, Greene (1999).
364 NOTES

Chapter 4
1. Runge translations by the author.
2. Grassmann translation by the author.
3. Helmholtz (1909, Vol II, p. 130).
4. Hering definition of constant veiling from Hering (1964, pp. 51–52).
5. Chromo-luminarisme, a term invented by the French painter George Seurat
(1859–1891) to designate his early style of neoimpressionist painting. The term
chromolithographe was first mentioned in French literature in 1837.
6. On Ostwald’s view of Helmholtz’s brightness definition, see Schwarz (1995).
7. Godlove formula from Judd (1969).
8. Pieter van Musschenbroek (1692–1761) was the inventor of the Leyden flask, a
form of electrical capacitor. The law of disk mixture was developed by Plateau in
1853 and the technique perfected by Maxwell (Boring, 1942).

Chapter 5
1. For a short history of photometry, see Walsh (1958).
2. On the Hefner lamp, see Walsh (1958).
3. Lambert comment in Lambert (1760).
4. On Treviranus and Boll, see Polyak (1957).
5. For a lively description on the CIE standard observer development, see Wright
(1996).
6. The density and distribution of cone types varies throughout the retina. The
macular spot has an irregular distribution and is absent in the central area of
focus of the normal eyes optics. To account for the average observer, for these
differences two different standard observers have been specified by the CIE, one
applying to a visual field subtending 2° and the other 10°.
7. For a trenchant critique of the CIE colorimetric system, see Cohen 2001. For a jab
see the comment by the eminent visual physiologist W. A. H. Rushton: “The CIE
triangle is brilliantly ingenious as an aid to the calculation of chromaticities which
can be upheld in a court of law where colour specification is in dispute. But the tri-
angle is monstrous as an indication of what is going on in the mechanism of vision.
It displays all colours as a mixture of three primary lights, none of which have an
existence that can be easily imagined. One of the three primaries is bright: it is pure
green from which is subtracted a lot of red, which it does not contain. The other
primaries are quite dark; they have strong colour but zero luminance. These do not
seem to me ingredients that lead to clarity in our conception of colour mechanisms
and I am astonished that some physiologists and many psychologists employ
them to instruct the young and bewilder the old.” (Journal of Physiology 1972;
220:178)
8. Many people have contributed to the elucidation of the genetic basis of color
vision. Among the pioneers were J. Nathans, R. and S. Yokoyama, and others. For
a succinct history, see Sharpe et al. (1999).
9. On Granit and Svaetichin, see Polyak (1957).
 NOTES 365

10. The notion of a central fovea free of S cones was first proposed by Artur
König in 1894. Since then it was confirmed in some experiments but remains
controversial.
11. The effect of field size on observed appearance in unpublished results by the
author.

Chapter 6
1. For an extended discussion on the concept of line element, see Wyszecki and Stiles
(1982, p. 654ff). See MacAdam (1981) on Schrödinger’s and Stiles’s line elements.
2. The terms NBS unit or judd have not gained widespread use.

Chapter 7
1. On Munsell system development, see Munsell (1918) and Kuehni (2002a).
2. The supplier of the Munsell Book of Colors is GretagMacbeth LLC, New Windsor,
NY.
3. On the committee experiments, see Judd and Nickerson (1967) and Judd (1955,
1957, 1965, 1967).
4. This was a forced choice experiment in which the observer could only answer in
one of two ways, “larger” or “smaller.”
5. MacAdam revision of OSA-UCS, personal communication by J. T. Luke.
6. Hering translations by Hurvich and Jameson; see Hering (1905–1911).
7. On the “beauty test for acceptance,” see Hård et al. (1996a).

Chapter 8
1. MacAdam’s ellipse 1 applies to a highly saturated reddish blue. It is more highly
saturated than any other color used in any color-matching error experiment. The
implicit S cone absorption value is very high and not in agreement with that of all
other ellipses. It has been left out of the analysis for that reason.
 Glossary

Acceptability Judgment of the perceived size of a color difference against an

internal standard of acceptability as a color match; used in color quality
control.
Achromatic color A perceived color without hue: white, gray, black.
Adaptation, visual Modification of the visual response to stimuli due to the
effects of the immediate surround and the total visual field of simultaneous
or preceding stimuli. There is brightness as well as chromatic adaptation.
Additive Produced by addition; specifically that the physical sum of two
visual stimuli is seen as the psychological sum in the sense of matching color
perceptions.
Aim color A color specification that is the target to be achieved by a color
chip, typically in a systematic collection.
Antagonistic Opposition in physiological action, specifically referring to
neurons with opponent color character.
Attribute An inherent characteristic; there are two sets of widely accepted
primary color attributes for object colors: (1) Hue, chroma, and lightness;
(2) hue, whiteness, and blackness.
Attribute measurement Process of assigning numbers or other symbols to
things in a manner that their relationship reflects the relationships of the
attribute being measured.
Bezold-Brücke effect A sensory effect named after German scientists,accord-
ing to which the hue sensation caused under normalized viewing conditions
by light of all but three wavelengths changes with changing intensity.

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.
366
 GLOSSARY 367

Blackness Degree of resemblance of a visual field to the fundamental color

contrast perception of black. A fundamental color attribute in the Hering
system.
Brightness Attribute of a visual perception according to which an area
appears to emit, or reflect, more or less light. Differences in brightness range
from bright to dim.
Chroma The attribute of a visual sensation permitting the judgment of the
degree to which a chromatic, related color differs from the achromatic color
of the same lightness.
Chromaticness Attribute of a visual sensation according to which the per-
ceived color of an area appears to be more or less chromatic.
Chromatic plane A plane in which all color perceptions, systematically
ordered, of colors seen as equally bright or light are located.
Chromaticity diagram A two-dimensional diagram in which colors can be
plotted according to their chromaticity coordinates, resulting in different
locations for colors of different hue and chromaticness.
CIE colorimetric system A color specification system developed by the
International Commission on Illumination (the acronym is derived from
the organization’s French name Commission Internationale de l’Éclairage).
Cleavage plane Cleavage is the tendency of crystalline materials to break
under strain along defined lines. A cleavage plane is a surface in a crys-
talline structure revealed after an actual or imagined break. In a color solid
it contains colors that stand in simple mathematically definable relationship
to each other.
Coefficient of variation A measure of the change in data; the standard devi-
ation of the data expressed as a percentage of the data mean.
Colorant A material that changes the absorption characteristics of other
materials: dyes or pigments, certain metal salts.
Color atlas A systematically arranged collection of colored chips that are
symbols of the colors of a color solid. The chips only illustrate the intended
space when viewed under prescribed conditions by an average color normal
observer.
Color appearance Appearance is the sense impression or aspect of a thing;
color appearance is the aspect of a colored field that distinguishes it from
the comparable aspect of another field that has a different color appear-
ance. Visual appearance includes visual aspects other than color, such as
glossiness, transparency, and opacity.
Color appearance models Mathematical models attempting to describe the
color appearance of objects as seen by the average observer under differ-
ent illumination and in different surrounds.
Color circle A circular arrangement of hues in their spectral order,
with nonspectral purple colors connecting the shortwave and the long
368 GLOSSARY

wave ends of the spectrum; usually illustrated with high chroma pigment
colorations.
Colorimeter Optical instrument for the investigation of color vision; in
technology also an instrument that measures the reflectance of materials
through three filters duplicating the color-matching functions of a standard
observer.
Colorimetry The branch of color science concerned with the numerical
specification of color stimuli.
Colorimetric purity A measure of saturation related to color stimuli and
expressed in the CIE chromaticity diagram. Its relationship to perceived
saturation in some standard conditions is complex.
Color difference The perceived difference between two non-identical fields
of color.
Color difference formula A mathematical formula that allows the calcula-
tion from stimuli of the difference between two color fields in a given
surround, as perceived by an average observer.
Color, full Translation of Hering’s term Vollfarbe, the mental image of a color
at its highest chromaticness; the color with a particular hue at the highest
level of chroma on the MacAdam limit.
Color harmony The combination of color elements in objects of art or craft
so that the effect is perceived as harmonious, in concord.
Color-matching error Stimulus variability in repeated matches of a standard
color.
Color-matching functions Three spectral functions describing the amounts
of three primary lights required to result in color perceptions matching
those obtained from spectral lights.
Color metric A metric describes the mathematical structure of a geometri-
cal space; a color metric applies to a color space, specifically a uniform color
space.
Color order Systematic arrangement of color perceptions in terms of
attributes and geometrical or mathematical models thereof.
Color, primary Colloquial term used in different circumstances: (1) One of
three lights whose color appearance cannot be matched by the other two
used with the other two to match the appearance of any other light; (2) one
of three colorants used in color order systems or in color reproduction, such
as yellow, red, and blue or yellow, magenta, and cyan; (3) one of the four
Hering Urfarben or fundamental hue perceptions of yellow, red, blue, and
green.
Color, related Color perception caused by light reflected from an object in
the presence of other objects. The perceived color depends on the perceived
color of surrounding objects.
Color, unrelated Color perceived to belong to an area seen in isolation from
other areas.
 GLOSSARY 369

Color solid Subset of color space containing, in a given experimental situa-

tion, all possible color experiences of the observer under consideration.
Color space Three-dimensional coordinate system within which color expe-
riences can be represented as points with unique positions. The term color
space should be limited to psychological spaces or psychophysical spaces
based on cone sensitivity or color-matching functions.
Color stimulus A stimulus is something that excites an organism, or one of
its components to functional activity. An external color stimulus normally
consists of light of one or more wavelengths, viewed against a surround of
different spectral composition.
Color zone theory Originally the merger of the Young-Helmholtz and the
Hering theory of color vision; more generally any model of color vision con-
sisting of two or more stages of processing.
Cones Cone-shaped light-sensitive cells in the retina. There are three types
of cones differing in spectral sensitivity in the normal human retina.
Cone sensitivity functions Spectral functions that describe the response of
the three cone types to light energy arriving at the surface of the retina.
Cone contrast diagram A diagram for illustrating the results of contrast
experiments in terms of cone activation, such as DM/M versus DL/L.
Contrast The difference between things having similar nature; specifically,
the degree of difference between two adjacent fields of color. Perceptually
contrast is expressed in terms of perceived difference, psychophysically in
cone activation (in a cone contrast diagram) or in colorimetric terms.
Correlation coefficient A number indicating the degree of association
between two sets of data.
Crispening Describes the fact that smallest increments in stimuli are neces-
sary for a criterion perceptual difference response if the surround color is
intermediate to the colors of the two fields compared, both in luminance
and chromaticity.
Criterion response Perceptual response at the level of the selected criterion;
the criterion is a standard on which the judgment is based.
Detection Discovery or determination of the existence or presence of some-
thing; specifically, for example, the determination of presence of redness in
a perceived color.
DeVries-Rose behavior Increase of the Weber fraction with the square root
of luminance instead of being constant.
Diapason The entire compass of musical notes.
Discrimination The process by which two stimuli differing in some aspect
result in different responses of some sort.
Dominant wavelength (of a color stimulus) Wavelength of the monochro-
matic stimulus that, when additively mixed with the appropriate amount of
achromatic stimulus, results in a color match with the test stimulus.
370 GLOSSARY

Empirical Originating in observation or experience.

Equal energy light source A theoretical light source that has a relative
spectral power distribution of 1.0 across the spectrum.
Euclidean Relating to, or based on the geometry of the Greek mathema-
tician Euclid; specifically that the three color attribute differences in a
complex difference sum as the square root of the sum of the squares of the
individual attribute differences.
Field of view The size of the retinal image expressed in solid angle. The CIE
has specified a 2° and a 10° standard observer.
Flicker Variation in brightness or hue perceived upon stimulation by
intermittent or temporally nonuniform light.
Fluorent Appearance of chromatic fields when their luminance or luminous
reflectance is higher than that of the surround, but not as much as to make
them appear luminous.
Fluorescence A form of luminescence, property of certain inorganic and
organic molecules to reemit absorbed ultraviolet or visible light energy in
the visible region of the spectrum.
Ganzfeld A situation in which the entire visual field is identical in com-
position. There are different degrees of ganzfeld mentioned in literature.
In one situation, the observer has her head in a uniformly light-emitting
sphere. It is still possible to see the nose and other facial features and con-
trast is thereby possible. In another case, the observer has, say, identical
colored ping-pong balls with a section removed taped to her eyelids so that
no facial or other contrasting feature can be seen. Here the ganzfeld is
complete.
Geodesic The shortest line between two points on a given surface. The cur-
vature of the line depends on the geometry of the space.
Gestalt psychology The study of perception and behavior based on the indi-
vidual’s response to configurational wholes, stressing the uniformity of the
psychological events and rejecting analysis into discrete aspect.
Grade A position in a scale of ranks or qualities; specifically a fixed point in
a color scale.
Gray scale A series of grades representing an achromatic color scale, usually
with visually equidistant steps between neighboring grades.
Helmholtz-Kohlrausch effect Describes the fact of heterochromatic bright-
ness matching that chromatic colors are perceived as brighter than achro-
matic colors of the same luminance. The effect is dependent on the
dominant wavelength of the color.
Heterochromatic Of mixed chromatic appearance.
Hue Attribute of a visual perception according to which an area appears to
be similar to one of the colors yellow, red, blue, or green or to a combina-
tion of adjacent pairs of these colors considered in a closed ring.
 GLOSSARY 371

Hues, unique The four hues of the color circle that can not be matched with
colors other than themselves; the psychological primary hues yellow, red,
blue, and green. Unique red is a red hue that is neither yellowish nor bluish,
for example.
Hue superimportance Refers to the fact that a smaller stimulus increment is
required for a criterion difference response if it represents a hue difference
than if it represents a chroma or saturation difference of the same perceived
magnitude.
Illuminant An illuminating device; technically a set of numbers representing
the spectral power distribution of a light source.
Isomorphism A one-to-one correspondence between mathematical sets;
specifically, mapping of objects of color experience to objects in a geomet-
rical space so that a one-to-one correspondence is obtained.
Just noticeable difference (JND) Threshold difference; the initial perceptual
difference that can be seen when one of two originally identical fields of
color changes in any given direction.
Lateral geniculate nucleus A mass of cells in the brain along the visual pas-
sageway between the retina and the visual area at the back of the brain.
Lattice A regular geometrical arrangement of points over an area or in a
space; specifically related to the arrangement of colors in a color space.
Lightness Perceptual attribute of related colors according to which a
color field appears to emit equal or less light compared to a white field.
Lightness can be understood as relative brightness.
Line element The first fundamental form of a regular surface. It is defined
by the Riemannian metric. In connection with color the term is used to
describe a certain kind of color space defined by (weighted) increments of
color fundamentals.
Linear model Of the first degree with respect to variables; having a graph
that is a straight line.
Linear regression A functional relationship between two or more variables
in which the variables are linearly related.
Linking proposition Postulated link between two sets of facts that are only
indirectly related.
Luminance Luminous flux of a light beam emanating from a surface in a
given direction, per unit solid angle.
Luminous reflectance Luminance of the surface of an object compared to
the luminance of the surface of a perfectly reflecting diffuser, illuminated
with the same light source and viewed at the same angle. Also known as
luminance factor Y.
Macula “Yellow spot,” an irregularly formed ring-like area of yellowish
pigment in the central region of the retina. The fovea is located in the
central area of the macular ring, free of macular pigment.
372 GLOSSARY

Magnitude, sensory A numerical or symbolic quantitative measure of the

result of a sensory perception.
Magnocellular Relates to layers in the lateral geniculate nucleus in which
relatively large cells are located believed to relay information necessary for
motion perception.
Masking The reduction or suppression of one percept by the presence of
another.
Maximal color See full color.
Metamers Two or more differing spectral power distributions resulting in
identical color perceptions for an observer. Also used for objects with dif-
ferent reflectance functions seen as having identical color when viewed in
standard conditions under a given light source.
Monochromatic Light of a single wavelength or a very narrow band of
wavelengths seen as having identical color.
Monolexemic Describes a word consisting of a single meaningful linguistic
unit.
Neuron Cell in the nervous system specialized in the transmission of
electrical signals.
Neurophysiology Organic processes and phenomena of the nervous system.
Object color Apparent color of an object. The color of an object can vary
depending on the surround and contextual conditions in which it is viewed.
Opponent color theory A theory according to which color perception is
based on unique hues forming opposing pairs: red-green, yellow-blue, as
well as the non-hued pair black and white.
Orthogonal To intersect or lie at right angles.
Parabolic Refers to a type of curved line resulting from slicing a cone at a
certain range of angles.
Parvocellular Relates to layers in the lateral geniculate nucleus in which rel-
atively small cells are located believed to relay information necessary for
brightness and color perception.
Perception The subjective, conscious awareness of any aspect of the exter-
nal or internal environment.
Photometry The measurement of light as related to the average human
observer.
Power law A mathematical, exponential relationship between two variables;
specifically between a physical stimulus and the perception resulting
from it.
Psychometric function Plots the relative frequency of judgments “smaller
than,” “equal,” and “larger than” relative to the magnitude of the stimulus.
Psychophysics The study of mental processes by quantitative methods;
specifically the reports of human subjects of the perceptions resulting from
carefully measured light stimuli.
 GLOSSARY 373

Reflection The process by which a smooth surface returns electromagnetic

radiation, specifically light. In reflection the radiation is returned by a
simple optical law: the angle of reflection equals the angle of incidence.
Relativize To treat or describe as not absolute or independent.
Retina A layer coating the inside of the camera type eye, containing the
light-sensitive rod and cone cells and cells connected to them. The retina is
continuous with the optical nerve.
Riemannian geometry Non-euclidean geometry with positive curvature in
which the parallel line postulate is replaced by the postulate that every pair
of straight lines intersects.
Rods Rod-shaped light-sensitive cells in the retina, specialized to operate
primarily at low light levels resulting in brightness perception only.
Root mean square error The square root of the arithmetic mean of the
squares of the deviations of the various items from the arithmetic mean of
the whole; also termed standard deviation.
Saturation Attribute of a visual perception which permits a judgment to be
made of the degree to which a chromatic stimulus differs in appearance
from that of an achromatic stimulus, regardless of their brightness.
Sensation Mental process due to bodily stimulation, now distinguished from
awareness of the result of the process.
Spectral spaces Spaces created from reflectance or spectral power distribu-
tion data by dimension reduction techniques other than those involving
color matching or cone sensitivity functions.
Spectrophotometer An instrument for measuring the relative intensities of
light in different spectral regions.
Stimulus An agent that directly influences the activity of a living organism
or one of its parts; specifically electromagnetic radiation within the visible
band.
Suprathreshold Exceeding the threshold; specifically a difference that is
larger than a threshold difference.
Symbolic A formal system of notation representing relationships.
Tetractys The pythagorean name for the sum of the first four integers
regarded as the source of all things.
Threshold Visual, the lowest level or increment of stimulus resulting in a
visual perception or a difference perception.
Trichromacy Relates to the theory that human color vision is based on the
activity of three cone types.
Tristimulus values The scalar values of the amounts of three primary lights
required to match a given light. The CIE tristimulus values X, Y, and Z refer
to non-real lights X, Y, and Z.
Tone A tint or shade of color, typically achieved by adding white and/or
black pigments to highly chromatic pigments.
374 GLOSSARY

Value Munsell’s term for the grades of a perceptually uniform gray scale.
Vision Process by which the extended visual system extracts information
from light energy to help generate appropriate response behavior.
Weber fraction Proportionality constant between the stimulus increment
and the absolute value of the stimulus.
Weighting To apply a statistically or otherwise determined weighting factor
to a variable.
Whiteness Attribute of a diffusing surface permitting, when viewed under a
standard light source, the judgment of similarity to a standard white surface
viewed in the same light.
 References

Abney, W. de W. 1910. On the change in hue of spectrum colors by dilution with white
light. Proceedings of the Royal Society (London) A83:120–127.
Abney, W. de W. 1914. Researches in Colour Vision and the Trichromatic Theory.
London: Longmans, Green.
Ackerman, J. S. 1980. On early Renaissance color theory and practice. In H. A. Millon
ed. Studies in Italian Art and Architecture, 15th through 18th Centuries. Cambridge:
MIT Press.
Adams, E. Q. 1923. A theory of color vision. Psychological Review 30:56–76.
Adams, E. Q. 1942. X-Z planes in the 1931 I.C.I. system of colorimetry. Journal of the
Optical Society of America 32:168–173.
Adams, E. Q., and Cobb, P. W. 1922. The effect on foveal vision of bright (and dark)
surroundings. Journal of Experimental Psychology 5:39–45.
Agoston, G. A. 1987. Color Theory and Its Application in Art and Design. Berlin:
Springer.
Aguilonius, F. 1613. Opticorum libri sex. Antwerp: Plantin (with illustrations by PP
Rubens).
Alberti, L. B. Leon Battista Alberti on painting, ed. J. Spenser. New Haven: New Haven
University Press, 1956.
Albertus Magnus. Opera omnia, 38 vols, ed. A. Borgnet. Paris: Vivès, 1890–99.
Allen, E., and Yuhas, B. 1984. Setting up acceptability tolerances: A case study. Color
Research and Application 9:37–48.
Anonymous (C.B.). 1708. Traité de la peinture en mignature. The Hague: van Dole.
Arago, D. F. J. 1850. Un movement modéré rend la visibilité des objets plus facile. In
Oeuvres complètes de François Arago, vol. 10. Paris: Gide, pp. 255–260.

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc.
375
376 REFERENCES

Aristotle. In J. Barnes, The complete works of Aristotle, vol 1. Princeton: Princeton

University Press, 1984.
ASTM E284. 1996. Standard Terminology of Appearance. West Conshohocken PA,
American Society of Testing Materials.
ASTM E1360. 1996. Standard Practice for Specifying Color by Using the Optical Society
of America Uniform Color Scales System. West Conshohocken PA, American
Society of Testing Materials.
Aubert, H. 1865. Physiologie der Netzhaut. Breslau: Morgenstern.
Aubert, H. 1876. Grundzüge der physiologischen Optik. Leipzig: Engelmann.
Averroës. Epitome of Aristotle’s Parva naturalia, transl. H. Blumberg. Cambridge, MA:
Medieval Academy of America, 1961.
Avicenna. Liber de anima, seu sextus de naturalibus, ed. S. van Riet. Leyden: Louvain,
1972.
Ayama, M., Nakatsue, T., and Kaiser, P. K. 1987. Constant hue loci of unique and binary
balanced hues at 10, 100, and 1000 Td. Journal of the Optical Society of America
A4:1136–1144.
Bacon, R. The “Opus Majus” of Roger Bacon, 3 vols., ed. J. H. Bridges. Oxford: Claren-
don Press, 1897.
Bacon, R. De multiplicatione specierum. In D. C. Lindberg, ed., Roger Bacon’s Philos-
ophy of Nature. Oxford: Clarendon Press, 1983.
Bacon, R. Opera hactenus inedita, ed. R. Steele. Oxford: Clarendon Press, 1937.
Badu, A. 1986. Large colour differences between surface colours. Ph.D. thesis. Univer-
sity of Bradford.
Baird, J. C. 1997. Sensation and Judgment. Mahwah, NJ: Erlbaum.
Balinkin, I. 1941. Measurement and designation of small color differences. Bulletin of
the American Ceramical Society 20:392–402.
Barasch, M. 1978. Light and Color in the Italian Renaissance Theory of Art. New York:
New York University Press.
Bäuml, K. H. 2002. Color appearance of spatial pattern: the role of increments and
decrements. Vision Research 42:1627–1637.
Bedford, R. E., and Wyszecki, G. 1958. Wavelength discrimination for point sources.
Journal of the Optical Society of America 48:129–135.
Bellamy, B. R., and Newhall, S. M. 1942. Attributive limens in selected regions of the
Munsell color solid. Journal of the Optical Society of America 32:456–473.
Benary, W. 1924. Beobachtungen zu einem Experiment über Helligkeitskontrast. Psy-
chologische Forschung 5:131–142.
Benson, W. 1868. Principles of the Science of Colours Concisely Stated to Aid and
Promote Their Useful Application in the Decorative Arts. London: Chapman and
Hall.
Berlin, B., and Kay, P. 1969. Basic Color Terms. Berkley CA: University of California
Press.
Bernoulli, D. 1744. Opera. Geneva: Cramer, (Exposition of a new theory on the mea-
surement of risk). Translation in Econometrica, 1954, 22:23–35.
Berns, R. S. 1993. The mathematical development of CIE TC 1-29 proposed color
difference equation. Proceedings of the 7th Congress of the International Colour
Association, Vol. B: C19-1–C19-4.
 REFERENCES 377

Berns, R. S. 1996. Deriving instrumental tolerances from pass-fail and colorimetric data.
Color Research and Application 21:459–472.
Berns, R. S. 2000. Billmeyer and Saltzman’s Principles of Color Technology, 3d ed. New
York: Wiley.
Berns, R. S. 2001. Derivation of a hue-angle dependent, hue-difference weighting func-
tion for CIEDE2000. Proceedings AIC Color 01. Bellingham WA: SPIE.
Berns, R. S., and Billmeyer, F. W. 1985. Development of the 1929 Munsell Book of
Color: A historical review. Color Research and Application 10:246–250.
Berns, R. S., Alman, D. H., Reniff, L., Snyder, G. D., and Balonon-Rosen, M. R. 1991.
Visual determination of supra-threshold color-difference tolerances using probit
analysis. Color Research and Application 16:297–316.
Bezold, W. von. 1874. Die Farbenlehre im Hinblick auf Kunst und Kunstgewerbe. Braun-
schweig: Westermann. American edition: The Theory of Color in Its Relation to Art
and Art-Industry. Boston: Prang, 1876.
Billmeyer, F. W., and Bencuya, A. K. 1987. Interrelation of Natural Color System and
the Munsell color order system. Color Research and Application 12:243–255.
Billock, V.A., Gleason, G. A., and Tsou, B. H. 2001. Perception of forbidden colors in
retinally stabilized images: an indication of softwired cortical color opponency?
Journal of the Optical Society of America A18:2398–2403.
Birren, F. 1987. M. E. Chevreul, The Principles of Harmony and Contrast of Colors and
Their Application to the Arts. With a special introduction and newly revised com-
mentary by Faber Birren. West Chester, PA: Schiffer.
Boll, F. 1877. Zur Anatomie und Physiologie der Retina. Archiv für Physiologie, p.4.
Boring, E. G. 1929. A History of Experimental Psychology. New York: Century.
Boring, E. G. 1939. Introduction to Psychology. New York: Wiley.
Boring, E. G. 1942. Sensation and Perception in the History of Experimental Psychol-
ogy. New York: Appleton.
Bouguer, P. 1729. Essai d’optique. Paris: Jombert.
Bouguer, P. 1760. Traité d’optique sur la gradation de la lumière. Paris: Guerin et
Delatour.
Boyle, R. 1664. Experiments and Considerations Touching Colours. London:
Herringman.
Boynton, R. M. 1983. Ten years of research with the minimally distinct border. In J. D.
Mollon and L.T. Sharpe, eds., Colour vision: Physiology and Psychophysics. London:
Academic Press, pp. 193–206.
Boynton, R. M., and Kambe, N. 1980. Chromatic difference steps of moderate size along
theoretically critical axes. Color Research and Application 5:13–24.
Boynton, R. M., Nagy, A. L., Olson, C. X. 1983. A flaw in equations for predicting
chromatic differences. Color Research and Application 8:
Breckenridge, F. C., and Schaub, W. R. 1939. Rectangular uniform-chromaticity-scale
coordinates. Journal of the Optical Society of America 29:370–377.
Brenner, E. 1680. Nomenclatura et species colorum. Stockholm.
Brewster, D. 1831. A Treatise on Optics. London: Longman, Rees.
Brill, M. H., Fairman, H. S., Hemmendinger, H., and Ladson, J. A. 2002. The color-order
screen book: A softcopy simulated color atlas with selectable observer and illumi-
nant. Proceedings AIC Color 01. Bellingham, WA: SPIE.
378 REFERENCES

Brücke, E. 1866. Die Physiologie der Farben für die Zwecke der Kunstgewerbe. Leipzig:
Hirzel.
Brown,W. R. J., and MacAdam, D. L. 1949.Visual sensitivities to combined chromaticity
and luminance differences. Journal of the Optical Society of America 39:808–834.
Brown, W. R. J. 1952. The effect of field size and chromatic surroundings on color dis-
crimination. Journal of the Optical Society of America 47:137–142.
Brown, W. R. J. 1957. Color discrimination of twelve observers. Journal of the Optical
Society of America 47:137–143.
Burnham, R. W. 1949. Comparison of color systems with respect to uniform visual
spacing. Journal of the Optical Society of America 39:387–392.
Burns, S. A., Cohen, J. B., and Kuznetsov, E. N. 1990. The Munsell color cystem in fun-
damental color space. Color Research and Application 15:25–51.
Cardanus, H. 1563. Hieronimi Cardani opera omnia, 10 vols. Lyon (vol. 2, p. 552ff).
Carroll, J. D., and Chang, J. J. 1970. Analysis of individual differences in multidimen-
sional scaling via an N-way generalization of “Eckhardt-Young” decomposition.
Psychometrika 35:283–319.
Castel, L. B. 1739. Projet d’une nouvelle optique des couleurs fondée sur les observa-
tions et uniquement relative à la teinture et les autres arts coloristes. Journal de
Trévoux, April.
Chalcidius. Platonis Timaeus interprete Chalcidio, ed. J. Wrobel. Leipzig: Teubner, 1876.
Chalmers, D. J. 1996. The Conscious Mind. New York: Oxford University Press.
Charpentier, A. 1888. La lumière et les couleurs au point de vue physiologique. Paris:
Baillière.
Cheung, M., and Rigg, B. 1986. Colour-difference ellipsoids for five CIE colour centres.
Color Research and Application 11:185–195.
Cennini, C. 1400. The craftsman’s handbook: “Il libro dell’ arte,” transl. D. V. Thomp-
son. New Haven: Yale University Press, 1933.
Chevreul, M. E. 1839. De la Loi du contraste simultané des couleurs. Paris: Pitois-
Levraux.
Chichilnisky, E. J., and Wandell, B. A. 1999. Trichromatic opponent color classification.
Vision Research 39:3444–3458.
Chou, W., Lin, H., Luo, M. R., Westland, S., Rigg, B., and Nobbs, J. 2001. Performance
of lightness difference formulae. Coloration Technology 117:19–29.
Chroma Cosmos 5000 1978. Tokyo: Japan Color Research Institute.
Chromaton 707 1982. Tokyo: Japan Color Research Institute.
CIE 1924. CIE Proceedings. Cambridge: Cambridge University Press, 1926.
CIE 1964 (U*V*W*)-System. 1963. CIE Proccedings. Vienna, Austria: Commission
Internationale de l’Éclairage.
CIE 1976 (L*u*v*)-System. 1986. No. 15.2, Colorimetry, 2nd ed. Vienna, Austria:
Commission Internationale de l’Éclairage.
CIE 1976 (L*a*b*)-System. 1986. No. 15.2, Colorimetry, 2nd ed. Vienna, Austria:
Commission Internationale de l’Éclairage.
CIE 1987. International lighting vocabulary, 4th ed. Vienna, Austria: Commission Inter-
nationale de l’Éclairage.
 REFERENCES 379

CIE 1990. CIE 86, 2° Spectral Luminous Efficiency Function for Photopic Vision.
Vienna: Commission Internationale de l’Éclairage.
Clark, F. J. J., McDonald, R., and Rigg, B. 1984. Modification to the JPC79 colour-
difference formula. Journal of the Society of Dyers and Colourists 100:128–132.
Cohen, J. B. 1964. Dependency of the reflectance curves of the Munsell color chips.
Psychonomic Science 1:369–370.
Cohen, J. B. 2001. Visual Color and Color Mixture. Urbana: University of Illinois
Press.
Cohen, J. B., and Kappauf, W. E. 1982. Metameric color stimuli, fundamental metamers,
and Wyszecki’s metameric blacks. American Journal of Psychology 95:537–564.
Cohn, J. 1894. Experimentelle Untersuchungen über die Gefühlsbetonung der Farben,
Helligkeiten und ihrer Combinationen. Philosophische Studien 10:562–602.
Craik, K. J. W. 1938. The effect of adaptation on differential brightness discrimination.
Journal of Physiology 92:406–421.
Crane, H. D., and Piantanida, T. P. 1983. On seeing reddish green and yellowish blue.
Science 221:1078–1080.
Cronin, T. W., and Marshall, N. J. 1989. A retina with at least ten spectral types of
photoreceptors in a stomatopod crustacean. Nature 339:137.
Cui, G., Luo, M. R., Rigg, B., Roesler, G., and Witt, K. 2002. Uniform colour spaces
based on the DIN99 colour-difference formula. Color Research and Application
27:282–290.
Davidson, H. R., and Friede, E. 1953. The size of acceptable color differences. Journal
of the Optical Society of America 43:581–589.
Delboeuf, J. R. L. 1873. Étude psychophysique. Mémoires couronnés et autres mémoires
publiés par l’Academie Royales des Sciences, des Lettres, et des Beaux-Arts de
Belgique, vol. 23. Brussels: Hayez.
Delboeuf, J. R. L. 1883. Examen critique de la loi psychophysique: sa base et sa signifi-
cation. Paris: Baillière.
De Matiello, M. L. F., Salinas, H., and Benavente, C. 2001. Dichoptic discrimination of
saturation. Color Research and Application 26 (suppl):S53–S56.
Democritus of Abdera. In Die Vorsokratiker II. Stuttgart: Reclam, 1986.
Dennett, D. C. 1991. Consciousness Explained. Boston: Little, Brown.
Derby, 1999. Available at ftp//colour.derby.ac.uk.
Derefeldt, G. 1991. Colour appearance systems. In P. Gouras, ed., The Perception of
Colour. Boca Raton, FL: CRC Press.
Derefeldt, G., and Sahlin, C. 1986. Transformation of NCS data into CIELAB colour
space. Color Research and Application 11:146–152.
Derefeldt, G., and Hedin, C. E. 1987. A color atlas for graphical displays. In B. Knave
and P. G. Wiedebäck, eds., Work with Display Units 86. Amsterdam: North-Holland.
Derrington, A. M., Krauskopf, J. and Lennie, P. 1984. Chromatic mechanisms in lateral
geniculate nucleus of macaque. Journal of Physiology 357:241–265.
De Valois, K. K., ed. 2000. Seeing. San Diego, CA: Academic Press.
De Valois, R. L. 1965. Analysis and coding of color vision in the primate visual system.
Cold Spring Harbor Symposium on Quantitative Biology 30:565–579.
380 REFERENCES

De Valois, R. L., and De Valois, K. K. 1993. A multi-stage color model. Vision Research
33:1053–1065.
De Valois, R. L., and De Valois, K. K. 1996. On “A three-stage color model.” Vision
Research 36:833–836.
De Valois, R. L., De Valois, K. K., Switkes, E., and Mahon, L. 1993. Hue scaling of
isoluminant and cone-specific lights. Vision Research 37:885–897.
De Valois, R. L., Cottaris, N. P., and Elfar, S. 1997. S-cone inputs to striate cortex cells.
Investigative Ophtalmology and Visual Science 38:S15.
DeVries, H. 1946. The luminosity curve of the eye as determined by measurements with
the flicker photometer. Physica 14:319–348.
Diderot, D. 1766. Essai sur la peinture. In Oeuvres complètes, 20 vols. Paris: P. Garnier
1875–1879.
DIN Farbenkarte, German Standard DIN 6164. 1962. Berlin: Beuth.
DIN 6176: Farbmetrische Bestimmung von Farbabständen bei Körperfarben nach der
DIN99 Formel. 2000. Berlin: Deutsches Institut für Normung.
Dobrowolsky, W. 1872. Beiträge zur physiologischen Optik. Archiv für Ophtalmologie
18:53–103.
Donders, F. C. 1881. Über Farbensysteme. Archiv für Ophtalmologie 26:155–223.
Doppler, C. 1848. Zwei weitere Abhandlungen aus dem Gebiete der Optik. 2. Versuch
einer systematischen Classification der Farben. Abhandlungen der königlichen
bömischen Gesellschaft der Wissenschaften 5(5):401–412. Reprinted in Schwarz, A.
1991/2. Christian Doopler’s Kugeloctant von 1847. Die Farbe 38:49–81.
Ebbinghaus, H. 1887. Die Gesetzmässigkeit des Helligkeitscontrastes. Sitzungsberichte
der Preussischen Akademie der Wissenschaften 2:995–1009.
Ebbinghaus, H. 1902. Grundzüge der Psychologie, Vol. 1. Leipzig: Viet.
Empedocles of Acragas. In Die Vorsokratiker II. Stuttgart: Reclam, 1986.
Eraclius. De coloribus et artibus romanorum. Original and translation in M. P. Merri-
field, de., Medieval and Renaissance Treatises on the Arts of Painting. New York:
Dover, 1967.
Eskew, R. T., McLellan, J. S., and Giulianini, F. 1999 Chromatic detection and discrim-
ination. In K. R. Gegenfurtner and L.T. Sharpe, eds., Color Vision. Cambridge:
Cambridge University Press.
Evans, R. M. 1974. The Perception of Color. New York: Wiley.
Evans, R. M., and Swenholt, B. K. 1968. Chromatic strengths of colors, Part II: The
Munsell System. Journal of the Optical Society of America 58:580–584.
Fechner, G. T. 1851. Zend-Avesta. Leipzig: Voss.
Fechner, G. T. 1860. Elemente der Psychophysik, transl. H.S. Langfeld. Leipzig:
Breitkopf und Härtel.
Farnsworth, D. 1944. The Farnsworth rectilinear uniform chromaticity diagram No. 38.
Memorandum 44-1. Medical Laboratory of the U.S. Submarine Base, New London,
CT.
Fairchild, M. D. 1998. Color Appearance Models. Reading, MA: Addison-Wesley.
Fairchild, M. D., and Pirotta, E. 1991. Predicting the lightness of chromatic object colors
using CIELAB. Color Research and Application 16:385–393.
 REFERENCES 381

Fechner, G. T. 1860. Elemente der Psychophysik, Vols. 1 and 2. Leipzig: Breitkopf und
Härtel. Translation: Elements of Psychophysics. Vol. 1, eds. E. G. Boring and D. H.
Howes. New York: Holt, Rinehart and Winston, 1966.
Feller, R. L., and Stenius, A. 1970. On the color space of Sigfrid Forsius, 1611. Color
Engineering 8:51–58.
Ficino M. 1519. Opera, I. Basel, pp. 825–826.
Field, G. 1835. Chromatography; or a Treatise on Colours and Pigments, and of their
Powers in Painting. London: Tilt.
Fiorentini, A., and Lee, B. B. 2000. Tobias Mayer’s On the relationship between colors.
Color Research and Application 25:66–74.
Fludd, R. 1629–31. Medicina catholica, 2 vols. Frankfurt: C. Rötelli.
Forsius, S. A. 1611. Physica, Codex Holmiensis D. 6, Royal Library, Stockholm.
Foss, C. E. 1947. Representations of color space and their applications. American
Ceramic Society Bulletin 27:55–63.
Foss, C. E. 1978. Space lattice used to sample the color space of the Committee on
Uniform Color Scales of the Optical Society of America. Journal of the Optical
Society of America 68:1616–1619.
Foss, C. E., Nickerson, D., and Granville, W. C. 1944. Analysis of the Ostwald color
system. Journal of the Optical Society of America 34:361–382.
Freeman, K. 1957. Ancilla to the pre-Socratic Philosophers. Cambridge: Harvard
University Press.
Friele, L. F. C. 1961. Analysis of the Brown and Brown-MacAdam colour discrimina-
tion data. Die Farbe 10:193–202.
Friele, L. F. C. 1965. Further analysis of colour discrimination data. Proceedings of the
International Colour Meeting (Lucerne 1965). Göttingen: Musterschmidt, 1966.
Friele, L. F. C. 1978. Fine color metric (FCM). Color Research and Application 3:53–64.
Frisch, J. C. 1788. Ueber eine harmonische Farben-Tonleiter, und die Wirkungen und
Verhältnisse der Farben im Colorite. Monatsschrift der Akademie der Künste und
mechanischen Wissenschaften zu Berlin 2(8):58–77.
Fullerton, G. S., and Cattell, J. K. 1892. On the Perception of Small Differences. Philadel-
phia: University of Pennsylvania Press.
Gage, J. 1993. Color and Culture. Boston: Little, Brown.
Gage, J. 1999. Color and Meaning. Berkley: University of California Press.
Gavel, J. 1979. Colour, A Study of its Position in the Art Theory of the Quattro- &
Cinquecento. Stockholm: Almquist and Wiksell.
Gelb, A. 1929. Die “Farbenkonstanz” der Sehdinge. In W. A. von Bethe, ed., Handbuch
der normalen und pathologischen Sehphysiologie, Vol 12:594–678. Berlin: Springer.
Gerritsen, F. 1984. Entwicklung der Farbenlehre. Göttingen: Muster-Schmidt.
Gerstner, K. 1986. The forms of color. Cambridge: MIT Press.
Gilchrist, A. L., ed. 1994. Lightness, Brightness, and Transparency. Hillsdale NJ:
Erlbaum.
Gilchrist, A. L., Kossyfidis, C., Agostini, T., Bonato, F., Cataliotti, J., Li, X., Spehar, B.,
Annan, V., and Economou, E. 1999. An anchoring theory of lightness perception.
Psychological Review 106:795–834.
382 REFERENCES

Gibson, K. S., and Nickerson, D. 1940. An analysis of the Munsell Color System based
on measurements made in 1919 and 1926. Journal of the Optical Society of America
30:591–608.
Gibson, K. S., and Tyndall, E. P. T. 1923. Visibility of radiant energy. Bulletin Bureau of
Standards 19:131.
Giulianini, F., and Eskew Jr., R. T. 1998. Chromatic masking in the (DL/L, DM/M) plane
of constant-contrast space reveals only two detection mechanisms. Vision Research
38:3913–3926.
Glasser, L. G., McKinney, A. H., Reilly, C. D., and Schnelle, P. D. 1958. Cube-root color
coordinate system. Journal of the Optical Society of America 48:736–740.
Glenn, J. T., and Killian, J. T. 1940. Trichromatic analysis of the Munsell Book of Color.
Journal of the Optical Society of America 30:609–616.
Glisson, F. 1677. Tractatus de ventriculo et intestinis. London: Brome.
Godlove, I. H. 1933. Neutral value scales: II. Comparison of results and equations
describing value scales. Journal of the Optical Society of America 23:419–425.
Godlove, I. H. 1951. Improved color-difference formula, with applications to the
perceptibility and acceptability of fadings. Journal of the Optical Society of America
41:760–772.
Goethe, J. W. v. 1810. Zur Farbenlehre. Tübingen: Cotta.
Granville, W. C. 1994. The Color Harmony Manual. Color Research and Application
19:77–98.
Grassmann, H. G. 1853. Zur Theorie der Farbenmischung, Poggendorff’s Annalen der
Physik und Chemie 29:69–84.
Green, D. M., and Swets, J. A. 1966. Signal Detection Theory and Psychophysics. New
York: Wiley.
Greene, B. 1999. The Elegant Universe. New York: Norton.
Guan, S. S., and Luo, M. R. 1999. A colour-difference formula for assessing large colour
differences. Color Research and Application 24: 344–355.
Guild, J. 1931. The colorimetric properties of the spectrum. Philosophical Transactions
of the Royal Society (London) A230:149–187.
Guilford, J. P. 1932. A generalized psychological law. Psychological Review 39:73–
85.
Guth, S. L. 1972. A new vector model. In J. J. Vos, L. F. C. Friele and P. L. Walraven,
eds., Color Metrics. Soesterberg: Institute for Perception TNO, pp. 82–98.
Guth, S. L. 1991. Model for color vision and light adaptation. Journal of the Optical
Society of America A8:976–993.
Guth, S. L. 2001. ATD01 model for color appearances and differences. Proceedings AIC
Color 01. Bellingham, WA: SPIE, 2002.
Hall, M. 1992. Color and Meaning. Cambridge: Cambridge University Press.
Hård, A., and Sivik, L. 1981. NCS-Natural Color System: A Swedish standard for color
notation. Color Research and Application 6:129–138.
Hård, A., Sivik, L., and Tonnquist, G. 1996. NCS, Natural Color System—From concept
to research and applications. Part I. Color Research and Applications 21:180–205.
Part II. Color Research and Applications 21:206–220.
Hardin, C. L. 1988. Color for Philosophers. Indianapolis IN: Hackett.
 REFERENCES 383

Harris, M. 1766. The Natural System of Colours. London: Laidler.

Hasegawa, I., and Miyashita, Y. 2002. Categorizing the world: Expert neurons look into
key features. Nature Neuroscience 5:90–91.
Hayter, C. 1826. A New Practical Treatise on the Three Primitive Colours. London:
printed for the author.
Helmholtz, H. v. 1856. Handbuch der physiologischen Optik, Teil I. Hamburg: Voss.
Helmholtz, H. v. 1860. Handbuch der physiologischen Optik, Teil II. Hamburg: Voss.
Helmholtz, H. v. 1866. Handbuch der physiologischen Optik, Teil III. Hamburg: Voss.
Helmholtz, H. v. 1891. Versuch einer erweiterten Anwendung des Fechnerschen Geset-
zes im Farbensystem. Zeitschrift für Psychologie und Physiologie der Sinnesorgane
2:1–30.
Helmholtz, H. v. 1896. Handbuch der physiologischen Optik, 2nd ed. Hamburg: Voss.
Helmholtz, H. v. 1909. Handbuch der physiologischen Optik, 3rd ed. Hamburg: Voss,
1909–1911. English translation: Helmholtz’s Treatise on Physiological Optics, transl.
from the third edition, J. P. C. Southhall ed. Washington: Optical Society of America,
1924.
Henry, C. 1888. Cercle chromatique et sensation des couleurs. La Revue Indépendante
(June):238–289.
Herbart, J. F. 1824–25. Die Psychologie als Wissenschaft. Königsberg: Unzer.
Hering, E. 1874. Zur Lehre vom Lichtsinn V. Grundzüge einer Lehre des Lichtsinnes.
Sitzungsberichte der Kaiserlichen Akademie der Wissenschaften in Wien. Mathe-
matisch-Naturwissenschaftliche Classe. Abtheilung III 69:85–104.
Hering, K. E. K. 1878. Zur Lehre vom Lichtsinne. Wien: Gerold.
Hering, E. 1905–11. Grundzüge der Lehre vom Lichtsinn. Berlin: Springer. Published
in installments 1905, 1907, 1911. English translation: Outlines of a Theory of the Light
Sense, transl. L. M. Hurvich and D. Jameson. Cambridge: Harvard University Press,
1964.
Hesselgren, S. 1952. Hesselgrens Färgatlas korfattad färglära. Stockholm: Palmer.
Hicketier, E. A. 1974. Color Atlas. New York: Van Nostrand Reinhold.
Höfler, A. 1897. Psychologie. Vienna: Tempsky.
Hunt, R. W. G. 1977. The specification of colour appearance. I. Concepts and terms.
Color Research and Application 2:55–68.
Hunter, R. S. 1942. Photoelectric Tristimulus Colorimetry with Three Filters. Circular
C429. National Bureau of Standards, Washington.
Hurvich, L. M. 1981. Color Vision. Sunderland: Sinauer.
Hurvich, L. M., and Jameson, D. 1955. Some quantitative aspects of an opponent-colors
theory. II. Brightness, saturation, and hue in normal and dichromatic vision. Journal
of the Optical Society of America 45:602–616.
Indow, T. 1988. Multidimensional studies of Munsell color solid. Psychological Review
95:456–470.
Indow, T. 1995. Psychophysical scaling: Scientific and practical applications. In R. D.
Luce et al, eds., Geometric Representations of Perceptual Phenomena, Mahwah NJ:
Erlbaum.
Indow, T. 1999a. Predictions based on Munsell notation: I. Perceptual color differences.
Color Research and Application 24:10–18.
384 REFERENCES

Indow, T. 1999b. Predictions based on Munsell notation: II. Principal hue components.
Color Research and Application 24:19–32.
Indow, T. 1999c. Principal hue curves and color difference. Color Research and Appli-
cation 24:266–279.
Indow, T. 2001. Uniformities in OSA-UCS and in NCS tested by color difference pre-
diction based on principal hue components. Proceedings AIC Color 01. Bellingham,
WA: SPIE, 2002.
Indow, T. 2002. Color differences predicted by color component differences. Color
Research and Application 27:425–429.
Indow, T., and Aoki, N. 1983. Multidimensional mapping of 178 Munsell colors. Color
Research and Application 8:145–152.
Indow, T., and Ohsumi, K. 1972. Multidimensional mapping of sixty Munsell colors by
non-metric procedure. In J. J. Vos, L. F. C. Friele and P. L. Walraven, eds., Color
Metrics. Soesterberg, Netherlands: Institute for Perception.
Indow, T., and Morrison, M. L. 1991. Construction of discrimination ellipsoids for
surface colors by the method of constant stimuli. Color Research and Application
16:42–56.
Ives, H. E. 1912. Studies in the photometry of lights of different colours: I. Spectral
luminosity curves obtained by the equality of brightness photometer and flicker
photometer under similar conditions. Philosophical Magazine Series 6, 24:149–188.
Jacobson, E., Granville, W. C., and Foss, C. E. 1942. Color Harmony Manual. Chicago:
Container Corporation of America.
Jaeckel, S. M. 1975. The HATRA data on colour differences and visual colour passing.
Journal of the Society of Dyers and Colourists 91:242.
Jain, A. K. 1972. Color distance and geodesics in color 3-space. Journal of the Optical
Society of America 62:1287–1291.
Jameson, D. 1972. Theoretical issues of color vision. In D. Jameson and L. M. Hurvich,
eds., Handbook of Sensory Physiology, Vol 7/4 Visual Psychophysics. Berlin:
Springer.
Jameson, D., and Hurvich, L. M. 1964. Theory of brightness and color contrast in human
vision. Vision Research 4:135–154.
Jameson, K. A., Highnote, M. S., and Wasserman, L. M. 2001. Richer color experience
in observers with multiple photopigment opsin genes. Psychonomic Bulletin and
Review 8:244–261.
Jastrow, J. 1888. Critique of psychophysical methods. American Journal of Psychology
1:271–309.
Johannsen, T. 1937. Färg. Stockholm: Lindfors.
Jones, L. A. 1917. The fundamental scale of pure hue and retinal sensibility to hue
differences. Journal of the Optical Society of America 1:63–77.
Jordan, C., and Mollon, J. D. 1993. A study of women heterozygous for color deficien-
cies. Vision Research 33:1495–1508.
Judd, D. B. 1932. Chromaticity sensibility to stimulus differences. Journal of the Optical
Society of America 22:72–108.
Judd, D. B. 1935. A Maxwell triangle yielding uniform chromaticity scales. Journal of
the Optical Society of America 25:24–35.
 REFERENCES 385

Judd, D. B. 1936. Estimation of chromaticity differences and nearest color temperature

on the standard (ICI) colorimetric coordinate system. Journal of the Optical Society
of America 26:421–426.
Judd, D. B. 1940. The Munsell color system. Journal of the Optical Society of America
30:574–645.
Judd, D. B. 1955. Progress report by Optical Society of America Committee on Uniform
Color Scales. Journal of the Optical Society of America 45:673–676.
Judd, D. B. 1957. Progress report by Optical Society of America Committee on Uniform
Color Scales. Journal of the Optical Society of America 47:336.
Judd, D. B. 1965. Progress report for OSA Committee on Uniform Color Scales. Die
Farbe 14:287–295.
Judd, D. B. 1967. Committee on Uniform Color Scales—Fourth progress report. Journal
of the Optical Society of America 57:557.
Judd, D. B. 1969. Ideal color space. Palette 29:25–31, 30:21–28, 31:23–29.
Judd, D. B., and Kelly, K. L. 1939. Method of designating colors. Journal of Research of
the National Bureau of Standards 23:355–381.
Judd, D. B., Scofield, F., and Hunter, R. S. 1941. A proposed method of designating color.
ASTM Bulletin (May):19–24.
Judd, D. B., MacAdam, D. L., and Wyszecki, G. 1964. Spectral distribution of typical
daylight as a function of correlated color temperature. Journal of the Optical Society
of America 54:1031–1040.
Judd, D. B., and Nickerson, N. 1967. One set of Munsell Re-renotations. National
Bureau of Standards Report 192693 of December 26, 1967. Washington: U.S.
Department of Commerce.
Judd, D. B., and Yonemura, G. T. 1970. CIE 1960 UCS diagram and the Müller theory
of color vision. Journal of Research of the National Bureau of Standards 74A:23–30.
Judd, D. B., and Wyszecki, G. 1975. Color in Business, Science, and Industry, 3d ed. New
York: Wiley.
Kaneko, T. 1964. A reconsideration of the Cobb-Judd lightness function. Acta Chro-
matica 1:103–110.
Katz, D. 1909. Die Erscheinungsweisen der Farben und ihre Beeinflussung durch die
individuelle Erfahrung. Zeitschrift für Psychologie, Abteilung 1, Ergänzungsband 7.
Leipzig: J.A. Barth. Translation: Katz, D. 1935. The World of Colour. London: Kegan,
Paul.
Kelly, K. L., and Judd, D. B. 1976. Color: Universal Language and Dictionary of Names.
Washington: U.S. Government Printing Office, National Bureau of Standards
Special Publication 440.
Kim, D. H., and Nobbs, J. H. 1997. New weighting functions for the weighted CIELAB
colour difference formula. Proceedings 8th Congress of the International Colour
Association, Vienna: CIE, pp. 446–449.
Kircher A. 1671. Ars magna lucis et umbrae. Amsterdam: Jansson and Waesberge.
Kirschmann, A. 1895. Color-saturation and its quantitative relations. American Journal
of Psychology 7:386–404.
Klotz M. 1816. Gründliche Farbenlehre. München: Zängl.
Kohlrausch, V. A. 1935. Zur Photometrie farbiger Lichter. Das Licht 5:259–275.
386 REFERENCES

König, A. 1886. Über die neuere Entwicklung von Thomas Young’s Farbentheorie (On
the more recent development of Thomas Young’s theory of color). In Gesammelte
Abhandlungen zur physiologischen Optik. Leipzig: Barth, 1903.
König, A. 1891. Ueber den Helligkeitswert der Spectralfarben bei verschiedener
absoluter Intensität. In Beiträge zur Psychologie und Physiologie der Sinnesorgane.
Hamburg: Voss, pp. 309–388.
König, A. 1892. Die Grundempfindungen in normalen und anomalen Farbensystemen
und ihre Intensitätsvertheilung im Spectrum (In Gemeinschaft mit Conrad
Dieterici). Zeitschrift für die Physiologie und Psychologie der Sinnesorgane
4:241–347.
König, A., and Dietrici, C. 1884, Über die Empfindlichkeit des normalen Auges für
Wellenlängenunterschiede des Lichtes. Annalen der Physik und Chemie 22:579–
589.
König, A., and Brodhun, E. 1888–89. Experimentelle Untersuchungen über die psy-
chophysische Fundamentalformel in Bezug auf den Gesichtssinn. Sitzungsberichte
der preussischen Akademie der Wissenschaften 1888:917–931; 1889:641–644.
König, A., and Dietrici, C. 1892. Die Grundempfindungen in normalen und anomalen
Farbsystemen und ihre Intensitätsvertheilung im Spectrum. Zeitschrift für Psy-
chologie und Physiologie der Sinnesorgane 4:241–347.
Krantz, D. H., Luce, R. D., Suppes, P., and Tversky, A. 1971. Foundations of Measure-
ment, Vol. 1: Additive and Polynomial Representations. New York: Academic Press,
1971. Vol. 2: Geometrical, Threshold, and Probabilistic Representations (Suppes,
Krantz, Luce, and Tversky. Academic Press, 1989). Vol. 3: Representation, Axiomati-
zation, and Invariance (Luce, Krantz, Suppes, and Tversky. Academic Press, 1990).
Krauskopf, J. 1999. Higher order color mechanisms. In K. R. Gegenfurtner and L. T.
Sharpe, eds. Color vision. Cambridge: Cambridge University Press.
Krauskopf, J., and Gegenfurtner, K. 1992. Color discrimination and adaptation. Vision
Research 32:2165–2175.
Kries, J. von. 1882. Die Gesichtsempfindungen und ihre Analyse. Archiv für Anatomie,
Physiologie und wissenschaftliche Medicin (Suppl. Physiologische Abteilung): 1–178.
Kuehni, R. G. 1971a. Acceptability contours and small color difference formulas.
Journal of Color and Appearance 1(1):30–35,42.
Kuehni, R. G. (chairman). 1971b. Acceptability contours of selected textile matches in
color space. Textile Chemist and Colorist 3:248–255.
Kuehni, R. G. 1972. Color difference and objective acceptability evaluation. Journal of
Color and Appearance 1(3):4–10,15.
Kuehni, R. G. 1975. Visual and instrumental determination of small color differences.
Journal of the Society of Dyers and Colourists 91:68–71.
Kuehni, R. G. 1982. Advances in color difference formulas. Color Research and Appli-
cation 7:19–23.
Kuehni, R. G. 1999. Towards an improved uniform color space. Color Research and
Application 24:253–265.
Kuehni, R. G. 2000a. Uniform color space modeled with cone responses. Color Research
and Application 25:56–63.
Kuehni, R. G. 2000b. A comparison of five color order systems. Color Research and
Application 25:123–131.
 REFERENCES 387

Kuehni, R. G. 2000c. A structural comparison of the Munsell Renotation and the OSA-
UCS uniform color systems. Color Research and Application 25:186–192.
Kuehni, R. G. 2000d. An opponent-color model for the Sanders-Wyszecki Helmholtz-
Kohlrausch effect data set. Color Research and Application 25:292–293.
Kuehni, R. G. 2001a. Determination of unique hues using Munsell color chips. Color
Research and Application 26:61–66.
Kuehni, R. G. 2001b. Analysis of five sets of color difference data. Color Research and
Application 26: 141–150.
Kuehni, R. G. 2001c. Color space and its divisions. Color Research and Application
26:209–222.
Kuehni, R. G. 2001d. From color matching error to large color differences. Color
Research and Application 26:384–393.
Kuehni, R. G. 2001e. “Uniform” color space is not homogeneous. Proceedings AIC
Color 01. Bellingham, WA: SPIE, 2002.
Kuehni, R. G. 2002. The early development of the Munsell system. Color Research and
Application 27:20–27.
Kuehni, R. G., and Marcus, R. T. 1979. An experiment in visual scaling of small color
differences. Color Research and Application 4:81–91.
Kuehni, R. G., and Stanziola, R. 2002. Francis Glisson’s color specification system of
1677. Color Research and Application 27:15–19.
Küppers, H. 1968. Das Grundgesetz der Farbenlehre. Köln: DuMont.
Lambert, J. H. 1760. Photometria, sive de mensura et gradibus luminis, colorum et umbra.
Augsburg.
Lambert, J. H. 1772. Beschreibung einer mit dem Caulauischen Wachse ausgemalten
Farbenpyramide. Berlin: Haude und Spener.
Lang, H. 1980. Tobias Mayers Abhandlung über die Verwandtschaft der Farben 1758.
Die Farbe 28:1–34.
Larimer, J., Krantz, D. H., and Cicerone, C. M. 1974. Opponent-process additivity. I.
Red/green equilibria. Vision Research 14:1127–1140.
Larimer, J., Krantz, D. H., and Cicerone, C. M. 1975. Opponent-process additivity. II.
Yellow/blue equilibria and nonlinear models. Vision Research 15:723–731.
Le Blon, J. C. 1725. Coloritto or the harmony of colouring in painting reduced to mechan-
ical practice. London.
Lee, B. B. 1998. Receptors, channels and color in primate retina. In W. G. K. Backhaus,
R. Kliegl, J. S. Werner, eds., Color Vision. Berlin: Walter de Gruyter.
LeGrand, Y. 1949. Les seuils différentiels de couleurs dans la theorie de Young. Revue
d’Optique 28:261–278. Translated in: Color Research and Application 1994;19:
296–309.
Lennie, P. 1999. Color coding in the cortex. In K. R. Gegenfurtner and L. T. Sharpe,
eds., Color Vision. Cambridge: Cambridge University Press.
Lenz, R., Österberg, M., Hiltunen, J., Jaaskelainen, T., and Parkkinen, J. 1996. Unsu-
pervised filtering of color spectra. Journal of the Optical Society of America A13:
1315–1324.
Long, J., and Luke, J. T. 2001. The New Munsell Student Color Set. Lebanon, PA: Maple
Press.
388 REFERENCES

Lotto, R. B., and Purves, D. 1999. The effect of color on brightness. Nature Neuroscience
2:1010–1014.
Lotto, R. B., and Purves, D. 2002. A rationale for the structure of color space. Trends
in Neuroscience 25:84–89..
Lübbe, E. 1999. Empfindungsgemässe Farbbeschreibung unter Berücksichtigung des
Umfeldes. Dissertation. Aachen: Shaker Verlag.
Lucretius. 1968. De rerum natura (The way things are). Transl. R. Humphries.
Bloomington: Indiana University Press.
Luke, J. T. 2002. Color Cleaver® software. Marketed by Michael Wilcox School of
Colour, Bristol, England.
Luneburg, R. K. 1950. The metric of binocular visual space. Journal of the Optical
Society of America 40:627–642.
Luther, R. 1927. Aus dem Gebiete der Farbreizmetrik. Zeitschrift für technische Physik
8:540–558.
Luo, M. R., and Rigg, B. 1986. Chromaticity-discrimination ellipses for surface colours.
Color Research and Application 11:25–42.
Luo, M. R., and Rigg, B. 1987. BFD (l:c) colour-difference formula, Part I – Develop-
ment of the formula. Journal of the Society of Dyers and Colourists 103:86–94. Part
II – Performance of the formula. Journal of the Society of Dyers and Colourists
103:126–132.
Luo, M. R., Cui, G., and Rigg, B. 2001. The development of the CIE 2000 colour-
difference formula: CIEDE2000. Color Research and Application 26:340–350.
MacAdam, D. L. 1935. Maximum visual efficiency of colored materials. Journal of the
Optical Society of America 25:361–367.
MacAdam, D. L. 1937. Projective transformations of I.C.I. color specifications. Journal
of the Optical Society of America 27:294–299.
MacAdam, D. L. 1942. Visual sensitivities to color differences in daylight. Journal of
the Optical Society of America 32:247–274.
MacAdam, D. L. 1947. Note on the number of distinct chromaticities. Journal of the
Optical Society of America 37:308–309.
MacAdam, D. L. 1965. Specification of color differences. Acta Chromatica 1:147–156.
MacAdam, D. L. 1974. Uniform color scales. Journal of the Optical Society of America
64:1691–1702.
MacAdam, D. L. 1978. Colorimetric data for samples of OSA uniform color scales.
Journal of the Optical Society of America 68:121–130.
MacAdam, D. L. 1981. Color Measurement. Berlin: Springer.
MacAdam, D. L. 1990. Redetermination of colors for uniform scales. Journal of the
Optical Society of America A7:113–115.
MacAdam, D. L. 1995. Letter to Joy Turner Luke of March 22, 1995, with permission
of the recipient.
Mahy, M., Van Eycken, L., and Oosterlinck, A. 1994. Evaluation of uniform color spaces
developed after the adoption of CIELAB and CIELUV. Color Research and
Application 19:105–121.
Maloney, L. T. 1999. Physics-based approaches to modeling surface color perception.
In K. R. Gegenfurtner and L. T. Sharpe, eds., Color Vision. Cambridge: Cambridge
University Press.
 REFERENCES 389

Mandelstamm, E. 1867. Beitrag zur Physiologie der Farben. Archiv für Ophtalmologie
13:399–406.
Mansfeld, J. 1986. Die Vorsokratiker I and II. Stuttgart: Reclam.
Marks, L. E. 1978. The Unity of the Senses: Interrelations among the Modalities. New
York: Academic Press.
Marks, L. E., and Algom, D. 1998. Psychophysical scaling. In M. H. Birnbaum, ed.,
Measurement, Judgment, and Decision Making. New York: Academic Press.
Masson, V. 1845. Étude de photométrie électrique. Annales de chimie et de physique
14:129–195.
Mausfeld, R. 1998. Color perception: from Grassmann codes to a dual code for object
and illumination colors. In W. G. K. Backhaus, R. Kliegl and J. S. Werner, eds., Color
Vision. Berlin: Walter de Gruyter.
Mayer, T. 1775. Opera inedita Tobiae Mayeri; ed. G. C. Lichtenberg. Göttingen.
Maxwell, J. C. 1856. On the theory of colours in relation to colour-blindness. Transac-
tions of the Royal Scottish Society of Arts 4:394–400.
Maxwell, J. C. 1860. Theory of compound colours, and the relations of the colours of
the spectrum. Proceedings Royal Society (London) 10:57–84.
McDonald, R. 1974. The effect of non-uniformity in the ANLAB colour space on the
interpretation of visual colour differences. Journal of the Society of Dyers and
Colourists 90:189–198.
McDonald, R. 1980. Industrial pass-fail colour matching: Part I. Preparation of visual
colour-matching data. Journal of the Society of Dyers and Colourists 96: 372–376.
Part II. Methods of fitting tolerance ellipsoids. Ibid. 96:418–433. Part III. Develop-
ment of pass-fail formula for use with instrumental measurement of colour differ-
ence. Ibid. 96:486–495.
McLaren, K. 1971. Multiple linear regression: A new technique for improving colour-
difference formulae. Color Metrics. Soesterberg: AIC, pp. 296–307.
McLaren, K. 1976. An introduction to instrumental shade passing and sorting and a
review of recent developments. Journal of the Society of Dyers and Colourists
92:317–326.
McLaughlin, B. P. 2002. Color, consciousness, and color consciousness. In Q.
Smith and A. Jokic, eds., Aspects of Consciousness. Oxford: Oxford University
Press.
McMahon, E. 1956. Leonardo da Vinci: Treatise on Painting Codex Urbinas Latinus
1270. Princeton: Princeton University Press.
Melgosa, M., Hita, E., Romero, J., and Jiménez del Barco, L. 1992. Some classical color
differences calculated with new formulas. Journal of the Optical Society of America
A9:1247–1254.
Melgosa, M., Hita, E., Poza, A. J., Alman, D. H., and Berns, R. S. 1997. Suprathreshold
color-difference ellipsoids for surface colors. Color Research and Application
22:148–155.
Melgosa, M., Rivas, M. J., Hita, E., and Viénot, F. 2000. Are we able to distinguish color
attributes? Color Research and Application 25:356–367.
Merrifield, M. P. 1967. Medieval and Renaissance Treatises on the Art of Painting. New
York: Dover.
Müller, A. E. 1963. Swiss Color Atlas 2541. Winterthur: Chromos.
390 REFERENCES

Müller, G. E. 1896–97. Zur Psychophysik der Gesichtsempfindungen. Zeitschrift für

Psychologie 10:1–82, 321–413. Ibid. 14:1–76, 161–196.
Müller, G. E. 1930. Über die Farbenempfindungen. Zeitschrift für Psychologie,
Ergänzungsband 17 and 18. Leipzig: Barth.
Munsell, A. H. 1905. A Color Notation. Boston: Ellis. Current edition available from
Gretag-Macbeth, New Windsor, NY.
Munsell, A. H. 1907. Atlas of the Munsell Color System. Malden, MA: Wadsworth-
Howland.
Munsell, A. H. 1915. Atlas of the Munsell Color System, 2nd ed. Malden: Wadsworth-
Howland.
Munsell, A. H. 1918. Color diary 1899–1918. Unpublished.
Munsell Book of Color. Baltimore: Munsell Color Company, 1929. Modern editions
available from Gretag-Macbeth, New Windsor, NY.
Munsell, A. E. O., Sloan, L. L., and Godlove, I. H. 1933. Neutral value scales: I. Munsell
neutral value scale. Journal of the Optical Society of America 23:394–411.
Murdoch, J. E. 1984. Album of Science, Antiquity and the Middle Ages. New York:
Scribner.
Murray, D. J. 1993. A perspective for viewing the history of psychophysics. Behavioral
and Brain Sciences 16:115–186.
Musschenbroek, P. van. 1768. Introductio ad philosophiam naturalem,Vol 2.Amsterdam.
Morley, D. I., Munn, R., and Billmeyer, F. W. 1975. Small and moderate colour
differences II: The Monley data. Journal of the Society of Dyers and Colourists
91:229–242.
Nagy, A. L., Eskew, R. T., and Boynton, R. M. 1987. Analysis of color-matching ellipses
in cone excitation space. Journal of the Optical Society of America A4:756–768.
Natural Colour System Atlas. 1979. Stockholm: Scandinavian Colour Institute.
Nayatani, Y. 1997. Simple estimation methods for the Helmholtz-Kohlrausch effect.
Color Research and Application 22:385–401.
Nayatani, Y. 1998. A colorimetric explanation of the Helmholtz-Kohlrausch effect.
Color Research and Application 23:374–378.
Nayatani, Y., Sobagaki, H., and Hashimoto, K. 1994. Existence of two kinds of
representations of the Helmholtz-Kohlrausch Effect: I. The experimental confirma-
tion. Color Research and Application 19:246–261. II.The models. Color Research and
Application 19:262–272.
Neitz, M., Kraft, T. W., and Neitz, J. 1998. Expression of L-cone pigment gene subtypes
in females. Vision Research 38:3221–3225.
Newton, I. 1670. The Optical Papers of Isaac Newton: Vol. 1. The Optical Lectures
1670–1672, transl. ed. A. E. Shapiro. Cambridge: Cambridge University Press, 1984.
Newton, I. 1704. Opticks. London: Smith and Walford.
Newhall, S. M. 1940. Preliminary report of the OSA subcommittee on the spacing of
the Munsell colors. Journal of the Optical Society of America 33:617–645.
Newhall, S. M., Nickerson, D., and Judd, D. B. 1943. Final report of the O.S.A. sub-
committee on the spacing of the Munsell colors. Journal of the Optical Society of
America 33:385–418.
Nickerson, D. 1936. The specification of color tolerances. Textile Research 6:509–514.
 REFERENCES 391

Nickerson, D. 1940. History of the Munsell color system and its scientific application.
Journal of the Optical Society of America 30:575–586.
Nickerson, D. 1969. History of the Munsell color system. Color Engineering 7(5):42–51.
Nickerson, D. 1976. History of the Munsell color system, company, and foundation.
Color Research and Application 1:7–10.
Nickerson, D. 1977. History of the OSA Committee on Uniform Color Scales. Optics
News, Winter 1977:8–17.
Nickerson, D. 1978. Munsell Renotations for samples of OSA uniform color scales.
Journal of the Optical Society of America 68:1343–1347.
Nickerson, D., and Newhall, S. M. 1943. A psychological color solid. Journal of the
Optical Society of America 33:419–422.
Nickerson, D., and Stultz, K. F. 1944. Color tolerance specification. Journal of the Optical
Society of America 34:550–570.
Norwich, K. H. 1993. Information, Sensation, and Perception. New York:Academic Press.
Nyberg, N. D. 1929. Zum Aufbau des Farbkörpers im Raume aller Lichtempfindungen.
Zeitschrift für technische Physik 52:406–411.
Oleari, C. 1999. Inter-observer comparison of color-matching functions. Color Research
and Application 24:177–184.
Oleari, C. 2001. Comparisons between color space scales, uniform-color-scale atlases,
and color-difference formulas. Color Research and Application 26:351–361.
O’Regan, J. K., and Noë, A. 2001. A sensorimotor account of vision and visual con-
sciousness. Behavioral and Brain Sciences 24:912–1031.
OSA Uniform Color Scales. 1977. Washington: Optical Society of America.
Osgood, C. E. 1953. Method and Theory in Experimental Psychology. New York: Oxford
University Press.
Ostwald, W. 1918. Die Farbenlehre: Vol. 1. Mathetische Farbenlehre. Leipzig: Unesma.
Ostwald, W. 1919. Der Farbkörper. Leipzig: Unesma.
Palmer, G. 1777. Theory of Colours and Vision. London: Leacroft.
Palmer, S. E. 1999. Vision Science, Photons to Phenomenology. Cambridge: MIT Press.
Parkhurst, C. 1990. Roger Bacon on color. In K. L. Selig and E. Sears, eds., The Verbal
and the Visual, New York: Italica Press.
Parkhurst, C., and Feller, R. L. 1982. Who invented the color wheel? Color Research
and Application 7:217–230.
Photius. Bibliothèque, 8 vols., transl. R. Henry. Paris: Société d’éditions Les Belles
Lettres, 1959–1991.
Plateau, J. A. F. 1872. Sur la mesure des sensations physiques, et sur la loi qui lie
l’intensité et ces sensations a la cause excitante. Bulletins de l’Académie Royales des
Sciences, des Lettres et des Beaux-Arts de Belgique 33:376–388.
Plato. The Collected Dialogues, eds. E. Hamilton and H. Cairns. Princeton: Princeton
University Press, 1961.
Pointer, M. R., and Attridge, G. G. 1997. Some aspects of the visual scaling of large
colour differences. Color Research and Application 22:298–307.
Pointer, M. R., and Attridge, G. G. 1998. The number of discernible colours. Color
Research and Application 23:52–54.
392 REFERENCES

Pointer, M. R., and Attridge, G. G. 2000. Some aspects of the visual scaling of large
colour differences–II. Color Research and Application 25:116–122.
Pokorny, J., Smith, V. C., and Starr, S. J. 1976. Variability of color mixture data: II The
effect of viewing field size on the unit coordinates. Vision Research 16:1095–1098.
Polyak, S. 1957. The Vertebrate Visual System. Chicago: University of Chicago Press.
Pope, A. 1929. The Painter’s Terms. New Haven, CT: Yale University Press.
Porphyry. Isagoge, Porphyry the Phoenician, trans. E. W. Warren. Toronto: Pontifical
Institute of Mediaeval Studies, 1975.
Priest, I. G., Gibson, K. S., and McNicholas, H. J. 1920. An examination of the Munsell
color system. I. Spectral and total reflection and the Munsell scale of value. U.S.
National Bureau of Standards Technical Paper 167.
Priest, I. G., and Brickwedde, F. G. 1938. The perceptible colorimetric purity as a
function of dominant wavelength. Journal of the Optical Society of America
28:133–139.
Purves, D., Lotto, B., and Polger, T. 2000. Color vision and the four-color-map problem.
Journal of Cognitive Neuroscience 12:233–237.
Purves, D., and Lotto, R. B. 2002. Why We See What We Do: Evidence for a Wholly
Probabilistic Strategy of Vision. Sunderland: Sinauer Associates.
Qiao, Y., Berns, R. S., Reniff, L., and Montag, E. 1998. Visual determination of hue
suprathreshold color-difference tolerances. Color Research and Application 23:302–
313.
Ramanath, R., Kuehni, R., Snyder, W. E., and Hinks, D. 2003. Spectral spaces and color
spaces. Color Research and Application 28: forthcoming.
Rich, R. M., Billmeyer, F. W., and Howe, W. G. 1975. Method for deriving color-
difference-perceptibility ellipses for surface-color samples. Journal of the Optical
Society of America 65:956–959.
Richter, K. 1985. Farbempfindungsmerkmal Elementarbuntton und Buntheitsabstände
als Funktion von Farbart und Leuchdichte von In- und Umfeld. Bremerhaven:
Wirtschaftsverlag.
Richter, K. 1980. Cube-root color spaces and chromatic adaptation. Color Research and
Application 5:25–43.
Richter, K. 1996. Computergrafik und Farbmetrik. Berlin: VDE Verlag.
Richter, M. 1976. Einführung in die Farbmetrik. Berlin: Walter de Gruiter.
Richter, M., and Witt, K. 1986. The story of the DIN color system. Color Research and
Application 11:138–145.
Ridgway, R. 1912. Color Standards and Color Nomenclature. Washington: published by
the author.
Robertson, A. R. 1978. CIE guidelines for coordinated research on colour-difference
evaluation. Color Research and Application 3:149–151.
Robinson, F. D. 1962. Acceptability of color matches. Journal of the Oil Colour Chemist
Association 52:15–45.
Rohner, E., and Rich, D. C. 1996. Eine angenähert gleichförmige Farbabstandsformel
für industrielle Farbtoleranzen von Körperfarben. Die Farbe 42:207–220.
Rood, N. O. 1879. Modern Chromatics. New York: Appleton.
Rösch, S. 1928. Zur Kennzeichnung der Farben. Physikalische Zeitschrift 29:83–91.
 REFERENCES 393

Runge, P. O. 1810. Die Farben-Kugel oder Construction des Verhältnisses aller

Mischungen der Farben zueinander. Hamburg: Perthes.
Russell, B. 1912. The Problems of Philosophy. London: Oxford University Press.
Sanders, C. L., and Wyszecki, G. 1964. Correlate for brightness in terms of CIE color
matching data. In CIE Proceedings 15th Session, Vienna, 1963. Paris: CIE Central
Bureau.
Sankeralli, M. J., and Mullen, K. T. 1996. Estimation of the L-, M-, and S-cone weights
of the postreceptoral detection mechanisms. Journal of the Optical Society of
America A13:906–915.
Sankeralli, M. J., and Mullen, K. T. 1999. Ratio model for suprathreshold hue-increment
detection. Journal of the Optical Society of America A16:2625–2637.
Sarle, W. S. 1995. Measurement theory: Frequently asked questions. In Disseminations
of the International Statistical Applications Institute, 4th ed., Wichita: ACG Press, pp.
61–66.
Saunders, B., and van Brakel, J. 2002. Theories, Technologies, Instrumentalities of Color.
Lanham MA: University Press of America.
Saunderson, J. L., and Milner, B. I. 1946. Modified chromatic value space. Journal of the
Optical Society of America 36:36–42.
Schiffermüller, I. 1771. Versuch eines Farbensystems. Vienna.
Schönfelder, W. 1933. Der Einfluss des Umfeldes auf die Sicherheit der Einstellung von
Farbgleichungen. Zeitschrift für Sinnesphysiologie 63:228–236.
Schrödinger, E. 1920. Grundlinien einer Theorie der Farbmetrik im Tagessehen.
Annalen der Physik 63:297–447, 481–520. (Outline of a theory of color measurement
for daylight vision.) Transl.: D. L. MacAdam, ed., Sources of Color Science. Cam-
bridge: MIT Press.
Schrödinger, E. 1926. Über das Verhältnis der Vierfarben-zur Dreifarbentheorie.
Sitzungsberichte der Kaiserlichen Akademie der Wissenschaften in Wien.
Mathematisch-naturwissenschaftliche Classe 134(2a):471–490. Transl.: National
Translation Center in Color Research and Application 1994;19:37–47.
Schultze, M. J. S. 1866. Über den gelben Fleck der Retina, seinen Einfluss auf normales
Sehen und auf Farbenblindheit. Bonn: Cohen.
Schultze, W., and Gall, L. 1971. Application of color difference formulas to highly
saturated colors differing only in lightness and saturation. Journal of Color and
Appearance 1(1):17–24.
Schwarz, A. 1992. Christian Doppler’s Kugeloctant von 1847–Ein vergessenes
Farbsystem. Die Farbe 38:49–81.
Schwarz, A. 1995. Physikalisch begründete Farbsysteme. Die Farbe 41:31–60.
Schwarz, A. 1999. Die Lehren von der Farbenharmonie (The doctrines of color
harmony). Göttingen: Muster-Schmidt.
Seim, T., and Valberg, A. 1986. Towards a uniform color space: A better formula to
describe the Munsell and OSA color scales. Color Research and Application 11:11–24.
Semmelroth, C. C. 1970. Prediction of lightness and brightness on different back-
grounds. Journal of the Optical Society of America 60:1685–1689.
Semmelroth, C. C. 1971. Adjustment of the Munsell-value and W*-scale to uniform
lightness steps for various background reflectances. Applied Optics 10:14–18.
394 REFERENCES

Sève, R. 1979. A bibliography on whiteness. Die Farbe 26:89–104.

Sharpe, L. T., Stockman, A., Jägle, H., Knau, H., and Nathans, J. 1999. L, M and L-M
hybrid cone photopigments in man: Deriving lmax from flicker photometric spectral
sensitivities. Vision Research 39:3513–3525.
Shepard, R. N. 1962. The analysis of proximities: multidimensional scaling with an
unknown distance function (parts 1 and 2). Psychometrica 27:125–140, 219–246.
Shepard, R. N. 1992. The perceptual organization of colors: An adaptation to regulari-
ties of the terrestrial world? In J. H. Barkow, L. Cosmides and J. Tooby, eds., Oxford:
The Adapted Mind. Oxford University Press.
Shepard, R. N. 1994. Perceptual-cognitive universals as reflections of the world.
Psychonomic Bulletin and Review 1:2–28.
Shinomori, K., Schefrin, B. E., and Werner, J. S. 1997. Spectral mechanisms of spatially-
induced blackness: Data and quantitative model. Journal of the Optical Society of
America A14:372–387.
Silberstein, L. 1942. A fundamental criterion of uniform representability of equilumi-
nous colors on a geometrical surface. Journal of the Optical Society of America
32:552–556.
Smith, V. C. 1999. Private communication.
Smith, V. C., and Pokorny, J. 1975. Spectral sensitivity of the foveal cone photopigments
between 400 and 500 nm. Vision Research 15:161–171.
Smith, V. C., Pokorny, J., and Sun, H. 2000. Chromatic contrast discrimination: Data
and prediction for stimuli varying in L and M cone excitation. Color Research and
Application 25:105–115.
Sowerby, J. 1809. A New Elucidation of Colours, Original, Prismatic, and Material.
London: Taylor.
Speranskaya, N. I. 1959. Determination of spectrum color co-ordinates for twenty-seven
normal observers. Optics and Spectroscopy 7:424–428.
Stevens, S. S. 1946. On the theory of scales of measurement. Science 103:677–680.
Stevens, S. S. 1953. On the brightness of lights and the loudness of sounds. Science 118:
576 (Abstract).
Stevens, S. S. 1975. Psychophysics. New York: Wiley.
Stevens, S. S., and Galanter, E. H. 1957. Ratio scales and category scales for a dozen
perceptual continua. Journal of Experimental Psychology 54:377–411.
Stiles, W. S. 1946. A modified Helmholtz line element in brightness-colour space. Pro-
ceedings of the Physical Society (London) 58:41–47.
Stiles, W. S., and Burch, J. M. 1955. Interim report to the Commission International de
L’Eclairage, Zurich 1955, on the National Physical Laboratory’s investigation of
colour-matching. Optica Acta 2:168–181.
Stockman, A., MacLeod, D. I. A., and Johnson, N. E. 1993. Spectral sensitivities of the
human cones. Journal of the Optical Society of America A10:2491–2521.
Stockman, A., and Sharpe, L. T. 1998. Human cone spectral sensitivities: a progress
report. Vision Research 38:3193–3206.
Stockman, A., and Sharpe, L. T. 1999. Cone spectral sensitivities and color matching. In
K. R. Gegenfurtner and L. T. Sharpe, eds., Color Vision. Cambridge: Cambridge Uni-
versity Press.
 REFERENCES 395

Strocka, D. 1971. Color difference formulas and visual acceptability. Applied Optics 10:
1308–1313.
Strohmeyer, C. F., Chaparro, A., Rodriguez, C., Chen, D., Hu, E., and Kronauer, R. E.
1998. Short-wave cone signal in the red-green detection mechanism. Vision Research
38:813–826.
Stumpf, C. 1917. Die Attribute der Gesichtsempfindungen. Abhandlungen der königlich
preussischen Akademie der Wissenschaften. Philosophisch-historische Klasse 8.
Berlin: Verlag der Königlichen Akademie der Wissenschaften.
Sugiyama, Y., and Fukuda, T. 1959. Comparison of different U.C.S. by using MacAdam
ellipses and Munsell scales. Journal of Applied Physics (Japan) 28:757–766.
Svaetichin, G. 1956. Spectral response curves from single cones. Acta Physiologica Scan-
dinavica 134:17–46.
Swedish Standard SS 01 91 03. 1982. CIE tristimulus values and chromaticity
coordinates for the colour samples in SS 01 91 02. Stockholm: Swedish Standards
Institution.
Takasaki, H. 1966. Lightness change of grays induced by change in reflectance of gray
background. Journal of the Optical Society of America 56:504–509.
Takasaki, H. 1967. Chromatic changes induced by changes in chromaticity of
background of constant lightness. Journal of the Optical Society of America 57:93–
96.
Teller, D. A., and Pugh, E. N. 1983. Linking propositions in color vision. In J. D. Mollon
and L. T. Sharpe, eds., Colour Vision: Physiology and Psychophysics, London:
Academic Press, pp. 577–589.
Telesio, A. 1529. Antonii Thylesii Cosentini libellus de coloribus. Venice.
Teufel, H. J., and Wehrhahn, C. 2000. Evidence for the contribution of S cones to the
detection of flicker brightness and red-green. Journal of the Optical Society of
America A17:994–1006.
Theophilus, The Various Arts, ed. and transl. C. R. Dodwell. Oxford: Clarendon Press,
1986.
Thomsen, K. 2000. A euclidean color space in high agreement with the CIE94 color
difference formula. Color Research and Application 25:64–65.
Thurner, K., and Walther, V. 1970. Untersuchungen zur Korrelation von Farbabstands-
bewertungen auf visuellen Wegen und über Farbabstandsformeln. Proceedings
of the International Color Meeting “Color 69,” Vol. 2, pp. 671–687. Göttingen:
Musterschmidt.
Thurstone, L. L. 1927. Psychophysical analysis. American Journal of Psychology
38:368–389.
Titchener, E. B. 1896. Outline of Psychology. New York: Macmillan.
Tonnquist, G. 1966. A comparison between symmetrical and equi-spaced hue circles.
Die Farbe 15:376–388.
Torgerson, W. S. 1958. Theory and Methods of Scaling. New York: Wiley.
Treviranus, G. R. 1828. Beiträge zur Anatomie und Physiologie der Sinneswerkzeuge des
Menschen und der Thiere. Bremen: Heyse.
Trezona, P. W., and Parkins, R. P. 1998. Derivation of the 1964 colorimetric standards.
Color Research and Application 23:221–225.
396 REFERENCES

Turner, R. S. 1994. In the Eye’s Mind. Princeton: Princeton University Press.

Unger, F. W. 1854. Disque chromharmonique pour servir a expliquer les règles de
l’harmonie des couleurs. Göttingen.
Urso von Salerno. De comixtionibus elementorum libellus, ed. W. Stürner. Stuttgart:
Stuttgart University Press, 1976.
Usui, S., Nakauchi, S., and Nakano, M. 1992. Reconstruction of Munsell color space
by a five-layer neural network. Journal of the Optical Society of America A9:516–
520.
Valberg, A. 2001. Unique hues: An old problem for a new generation. Vision Research
41:1645–1657.
Veckenstedt, E. 1888. Geschichte der griechischen Farbenlehre. Paderborn: Schöningh.
Villalobos-Dominguez, C., and Villalobos, J. 1947. Atlas de los colores. Buenos Aires:
Libreria El Ateneo editorial.
Volbrecht, V. J., and Kliegl, R. 1998. The perception of blackness. In W. G. K. Backhaus,
R. Kliegl and J. S. Werner, eds., Color Vision. Berlin: De Gruyter.
Völz, H. G. 1998a. Die Berechnung grosser Farbabstände in nichteuklidischen
Farbräumen. Die Farbe 44:1–45.
Völz, H. G. 1998b. Transformation der CIE94-Formel in einen euklidischen Farbraum.
Die Farbe 44:97–105.
Völz, H. G. 1999. Die Euklidisierung des CMC-Raumes zur Berechnung grosser
Farbabstände. Die Farbe 45:1–23.
Vos, J. J. 1978. Colorimetric and photometric properties of a 2° fundamental observer.
Color Research and Application 3:125–128.
Vos, J. J. 1979. Line elements and physiological models of color vision. Color Research
and Application 4:208–216.
Vos, J. J., and Walraven, P. J. 1972. An analytical description of the line element in
the zone-fluctuation model of colour vision. Vision Research 12:1327–1344, 1345–
1366.
Vos, J. J., and Walraven, P. J. 1991. Back to Helmholtz. Color Research and Application
16:355–359.
Wachtler, T., Lee, T., and Sejnowski, T. J. 2001. Chromatic structure of natural scenes.
Journal of the Optical Society of America A18:65–77.
Wallach, H. 1948. Brightness constancy and the nature of achromatic colors. Journal of
Experimental Psychology 38:310–324.
Waller, R. 1686. A catalogue of simple and mixt colours, with a specimen of each colour
prefixt to its proper name. Philosophical Transactions of the Royal Society 26:24:32.
Walls, G. 1956. The G. Palmer story. Journal of the History of Medicine 11:66–96.
Walraven, P. L., and Bouman, M. A. 1966. Fluctuation theory of colour discrimination
of normal trichromats. Vision Research 6:567–586.
Walsh, J. T. W. 1958. Photometry, 3rd ed. London: Constable. Reprinted by Dover,
New York, 1965.
Wässle, H., Grünert, U., Martin, P. R., and Boycott, B. B. 1994. Immunocytochemical
characterization and spatial distribution of midget bipolar cells in the macaque
monkey retina. Vision Research 34:561–579.
Weber, E. H. 1834. De pulsu, resorptione, auditu et tactu. Leipzig: Koehler.
 REFERENCES 397

Webster, M. A., Miyahara, E., Malkoc, G., and Raker, V. E. 2000. Variations in normal
color vision. II. Unique hues. Journal of the Optical Society of America A17:1545–
1555.
White, M. 1981. The effect of the nature of the surround on the perceived lightness of
gray bars within square-wave test gratings. Perception 10:215–230.
Whittle, P. 1992. Whiteness, discriminability and the “crispening” effect. Vision Research
32:1493–1507.
Wiesel, T. N., and Hubel, D. H. 1966. Spatial and chromatic interactions in the lateral
geniculate body of the rhesus monkey. Journal of Neurophysiology 29:1115–1156.
Witt, K. 1981. Der Farbenraum nach DIN6164 in neuer Sicht. Die Farbe 29:17–52.
Witt, K. 1987. Three-dimensional threshold of color-difference perceptibility in painted
samples: variability of observers in four CIE color regions. Color Research and
Application 12:128–134.
Witt, K. 1990. Parametric effects on surface color-difference evaluation at threshold.
Color Research and Application 15:189–199.
Witt, K. 1999. Geometric relations between scales of small colour differences. Color
Research and Application 24:78–92.
Wright, W. D. 1928–29. A re-determination of the mixture curves of the spectrum.
Transactions of the Optical Society London 30:141–164.
Wright, W. D., and Pitt, F. H. G. 1934. Hue discrimination on normal colour-vision.
Proceedings of the Physical Society (London) 46:459–454.
Wright, W. D. 1941. The sensitivity of the eye to small colour differences. Proceedings
of the Physical Society (London) 53:93–112.
Wright, W. D. 1969. The Measurement of Colour, 4th ed. New York: Van Nostrand
Reinhold.
Wright, W. D. 1996. The origins of the 1939 CIE system. In P. K. Kaiser and
R. M. Boynton, Color Vision, second ed., Washington, DC: Optical Society of
America.
Wuerger, S. M., Maloney, L. T., and Krauskopf, J. 1995. Proximity judgments in color
space: Test of a euclidean color geometry. Vision Research 35:827–835.
Wuerschmidt, J. 1914. Dietrich von Freiberg, Ueber den Regenbogen. In C. Bauemker,
ed., Beiträge zur Geschichte der Philosophie des Mittelalters, Vol. 12. Münster:
Aschendorff.
Wundt, W. 1874. Grundzüge der physiologischen Psychologie. Leipzig: Engelmann.
Wundt, W. M. 1880. Grundzüge der physiologischen Psychologie. Leipzig: Engelmann.
Wundt, W. 1892. Vorlesungen ueber die Menschen- und Thierseele, zweite Auflage.
Leipzig: Voss.
Wundt, W. 1896. Grundriss der Psychologie. Leipzig: Alfred Körner.
Wyszecki, G. 1953. Valenzmetrische Untersuchung des Zusammenhanges zwischen
normaler und anomaler Trichromasie. Die Farbe 2:39–52.
Wyszecki, G. 1954. A regular rhombohedral lattice sampling of Munsell renotation
space. Journal of the Optical Society of America 44:725–734.
Wyszecki, G. 1960. Farbsysteme. Göttingen: Musterschmidt-Verlag.
Wyszecki, G. 1963. Proposal for a new color-difference formula. Journal of the Optical
Society of America 53:1318–1319.
398 REFERENCES

Wyszecki, G. 1967. Correlate for brightness in terms of CIE chromaticity coordinates

and luminous reflectance. Journal of the Optical Society of America 57:254–257.
Wyszecki, G. 1972. Recent developments on color-difference evaluation. In J. J. Vos, L.
F. C. Friele and P. L. Walraven, eds., Color Metrics. Soesterberg, Netherlands: Insti-
tute for Perception.
Wyszecki, G. 1981. Uniform color spaces. In Golden Jubilee of Colour in the CIE.
Bradford: Society of Dyers and Colourists.
Wyszecki, G., and Sanders, C. L. 1964. Correlate for lightness in terms of CIE-
tristimulus values: Part II. Journal of the Optical Society of America 47:840–842.
Wyszecki, G., and Wright, H. 1965. Field trial of the 1964 CIE color-difference formula.
Journal of the Optical Society of America. 55:1166–1174.
Wyszecki, G., and Fielder, G. H. 1971a. New color matching ellipses. Journal of the
Optical Society of America 61:1135–1152.
Wyszecki, G., and Fielder, G. H. 1971b. Color-difference matches. Journal of the Optical
Society of America 61:1501–1513.
Wyszecki, G., and Stiles, W. S. 1982. Color Science, 2nd ed. New York: Wileya
Xenophanes of Colophon. In K. Freeman, Ancilla to the Pre-Socratic Philosophers,
Cambridge: Harvard University Press, 1958.
Yamauchi, Y., Williams, D. R., Brainard, D. H., Roorda, A., Carroll, J., Neitz, M., Neitz,
J., Calderone, J. B., and Jacobs, G. H. 2002. What determines unique yellow, L/M cone
ratio or experience? Proceedings AIC Color 01. Bellingham, WA: SPIE.
Young, T. 1824. Chromatics. In supplement to the fourth, fifth, and sixth editions of the
Encyclopedia Brittanica. Edinburgh: A. Constable.
Zeki, S., and Bartels A. 1999. Toward a theory of visual consciousness. Consciousness
and Cognition 8:225–259.
 Credits

Permission to reproduce certain figures and text was obtained from the fol-
lowing copyright holders and is gratefully acknowledged:

American Psychological Association, Washington, DC:

Figure 4-4 reprinted from Psychological Review 95, 1988, p. 467, T. Indow, Mul-
tidimensional studies of Munsell color solid, with permission of the American
Psychological Association.

Cambridge University Press, Cambridge, UK

Figures 6-21 and 6-22 reprinted from Journal of Physiology 357, 1984, p. 244,
A. M. Derrington, J. Krauskopf and P. Lennie, Chromatic mechanisms in the
lateral geniculate nucleus of macaque, with permission of the senior author
Dr. A. M. Derrington, Nottingham, England.

Elsevier Science Inc. New York, NY:

Figure 6-23 reprinted from Vision Research 36, 1996, p. 834, R. L. De Valois
and K. De Valois, On “A three stage color model.”
Figure 6-24 reprinted from Vision Research 33, 1993, p. 1059, R. L. De Valois
and K. De Valois, A multi-stage color model.
Figures 6-25 and 6-26 reprinted from Vision Research 39, 1999, pp. 3446
and 3456, E. J. Chichilniski and B. Wandell, Trichromatic opponent color
classification.

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc. 399
400 CREDITS

Figure 8-12: reprinted from Vision Research 32, 1992, p. 2171, J. Krauskopf and
K. R. Gegenfurtner, Color discrimination and adaptation. All materials used
with permission from Elsevier Science Inc.

Gretag-Macbeth LLC, New Windsor, NY:

Figures 2-41, 2-42, 4-11 courtesy Gretag-Macbeth LLC.

Walter de Gruyter GmbH, Berlin, Germany:

Figure 2-48, reprinted from M. Richter, Einführung in die Farbmetrik 1976, p.
168, with permission of the publisher.

Dr. L. M. Hurvich, North Truro, MA:

Figures 5-13, 6-13, 6-14, 6-15, reprinted from L. M. Hurvich, Color Vision, 1981,
Sunderland, MA: Sinauer Associates, pp. 67, 71, 84, and 95, with permission of
the author.

MIT Press, Cambridge, MA:

Figures 2-49 and 2-50 reprinted from K. Gerstner, The Forms of Color,
Cambridge, MA: The MIT Press, 1986, p. 15, with permission of the publisher.

NCS Colour Center AB, Stockholm, Sweden:

Figures 2-52, 2-53, reprinted with permission from NCS Colour Center.

Optical Society of America, Washington, DC:

Figures 1-2 and 1-3 reprinted from Journal of the Optical Society of America
33 (1943) p. 420/21, D. Nickerson and S. M. Newhall, A psychological color
solid.
Figure 4-5 reprinted from Journal of the Optical Society of America A 9 (1992)
p. 518, S. Usui, S. Nakauchi, and M. Nakano, Reconstruction of Munsell color
space by a five-layer neural network.
Figure 4-13 reprinted from Journal of the Optical Society of America 58 (1968)
p. 581, R. M. Evans and B. K. Swenholt, Chromatic strength of colors, Part II:
The Munsell system.
Figure 5-12 reprinted from Journal of the Optical Society of America 32 (1942)
p. 172, E. Q. Adams, X-Z planes in the 1931 ICI system of colorimetry.
Figure 6-2 reprinted from Journal of the Optical Society of America 22 (1932)
p. 96, D. B. Judd, Chromaticity sensibility to stimulus differences.
Figure 6-3 reprinted from Journal of the Optical Society of America 25
(1935) p. 31, D. B. Judd, A Maxwell triangle yielding uniform chromaticity
scales.
Figure 6-4 reprinted from Journal of the Optical Society of America 26
(1936) p. 424, D. B. Judd, Estimation of chromaticity differences and nearest
color temperature on the standard 1931 (ICI) colorimetric coordinate system.
 CREDITS 401

Figure 6-28 reprinted from Journal of the Optical Society of America A 13

(1996) p. 1317. R. Lenz, M. Osterberg, J. Hiltunen, T. Jaaskelainen, and J.
Parkkinen, Unsupervised filtering of color spectra.
Figure 7-21 reprinted from Journal of the Optical Society of America 64
(1974), p. 1696, D. L. MacAdam, Uniform color scales.
Figure 7-23 reprinted from Journal of the Optical Society of America 68
(1978), p. 1616, C. E. Foss, Space lattice used to sample the color space of the
Committee on Uniform Color Scales. All materials used with permission of
the Optical Society of America.

Princeton University Press, Princeton, NJ:

Quotations of Aristotle in chapter 2 from The complete works of Aristotle,
Volume One, copyright 1984 by Jowett Copyright Trustees, reprinted with
permission of Princeton University Press.

Psychonomic Bulletin and Review, Austin, TX:

Figure 3-6: reprinted with permission from R. N. Shepard, Perceptual-cogni-
tive universals as reflections of the world, Psychonomic Bulletin and Review
1 (1994) p. 24.

Springer-Verlag, Heidelberg, Germany:

Figures 6-1, 6-7, 6-8, 6-9 reprinted from D. L. Mac Adam, Color Measurement,
1981, pp. 131, 132, 133, 153.
Figure 7-14 reprinted from G. A. Agoston, Color Theory and Its Application
in Art and Design, 1987, p. 223, with permission of Springer-Verlag.

John Wiley & Sons, Inc., New York, NY:

Figures 2-46, 2-47, 5-2, 5-6, 5-7, 5-18, 5-20, 6-17, 6-18, 7-9 reprinted from G.
Wyszecki and W. S. Stiles, Color Science, second edition, 1982, pp. 184, 257, 137,
499, 414, 166, 167, 510 respectively.
Figure 3-5 reprinted from S. S. Stevens, Psychophysics, 1975, p. 185.
Figure 4-33 reprinted from R. M. Evans, The Perception of Color, 1974,
p. 104.
Figure 4-9 reprinted from Color Research and Application 24 (1999) p. 270.
T. Indow, Principal hue curves and color difference.
Figure 5-27 reprinted from Color Research and Application 22 (1997) p. 155,
M. Melgosa, E. Hita, A. J. Poza, and R. S. Berns, Suprathreshold color-
difference ellipsoids for surface colors.
Figures 6-5, 6-30, 6-31 reprinted from Color Research and Application 19
(1994) p. 113, M. Mahy, L. Van Eycken, and A. Oosterlinck, Evaluation of
uniform color spaces developed after the adoption of CIELAB and CIELUV.
Figure 6-11 reprinted from Color Research and Application 11 (1986) p. 30,
M. R. Luo, and B. Rigg, Chromaticity discrimination ellipses for surface colors.
402 CREDITS

Figure 6-12 reprinted from Color Research and Application 26 (2001) p. 346,
M. R. Luo, G. Cui, and B. Rigg, The development of the CIE 2000 colour-
difference formula.
Figures 6-19 and 6-20 reprinted from Color Research and Application 15
(1990) p. 33, S. A. Burns, J. B. Cohen, and E. N. Kuznetsov, The Munsell color
system in fundamental color space.
Figures 7-26 and 7-27 reprinted from Color Research and Application 21
(1996) pp. 189 and 194 respectively, A. Hård, L. Sivik, and G. Tonnquist, NCS
Natural Color System, Part I.
Figure 7-28 reprinted from Color Research and Application 11 (1986) p. 147.
G. Derefeldt and C. Sahlin, Transformation of NCS data into CIELAB colour
space. All materials used with permission of John Wiley & Sons, Inc.
 Index

Abney, W. deW., 161 Boethius, 36–37

Abney effect, 284 Boring, E. G., 91
Acceptability, 228, 366 Bouguer, Pierre, 63, 152, 159
Adams, E. Q., 170, 214, 250 Boyle, Robert, 51
chromance diagram, 214 Breckenridge & Schaub diagram,
Adams-Cobb lightness scale, 177 208–209
Adams-Nickerson (ANLAB) formula, Brenner, Elias, 42
215, 226 Brewster, David, 72
Adaptation, 4, 366 Brightness, 111, 131–132, 340
Adobe Illustrator, Photoshop, 100 heterochromatic, 183, 341
d’Aguilon, François (Aguilonius), 37, 159 Brücke, E. W., 59, 126
Alberti, Leon Baptista, 32, 124 Burch, J. M., 163
Albertus Magnus, 30, 32
Anchoring problem, 153 Cardanus, Hieronimus, 34
Aristotle, 22ff, 25 Castel, L.-B., 47
Aubert, H., 76, 78 Categorization, 108
Averroës, 27 Cattell, J. M., 83
Avicenna, 23, 32 C. B., 46–47, 50
“Celtic” cross, 197–198
Bacon, Roger, 28, 30–32 Cennini, Cennino, 32, 124
Bartholomaeus Anglicus, 31 Chalcidius, 27, 31, 124
Benary effect, 154 Chalmers, D. J., 105
Benson, William, 78 Chevreul, M.-E., 59, 66–69
Berlin & Kay, 26, 109 Chroma, 21, 86, 128, 132, 147, 186–191,
Bezold, Wilhelm von, 78 331, 351, 356
BFD formula, 239–240, 268 circles, constant, 186-189
Blackness, 40, 76, 89, 97, 109, 129–130, Chroma Cosmos 5000, 86
304 Chromaticness, 9, 97

Color Space and Its Divisions: Color Order from Antiquity to the Present, by Rolf G. Kuehni
ISBN 0-471-32670-4 Copyright © 2003 by John Wiley & Sons, Inc. 403
404 INDEX

Chromaton 707, 86 opponent, 169–174

CIE, 91–92 personal, 3, 4
chromatic system, 93, 367 regular, 8–9, 101, 264, 336–337
chromaticity diagram, 164 uniform, 8ff, 14, 357–358
standard observers, 161 Color specification, 40, 90–91, 307
U*W*W* system, 211 Colorimeter, 161, 183, 216, 224, 334
CIE94 formula, 241–242 Colorimetric purity, 131, 318
optimization, 244–245 Colorimetric system, 161
CIEDE2000 formula, 245–347, 270 Cone contrast, 261, 312, 350, 369
CIELAB, 14, 101, 199, 230–231, 247, 268 Cone functions, 166–168
weights, 242–243 variation in, 168
Cleavage planes, 296 Cone response increments, 320–321
CMC formula, 235–236, 268 Cones, 4, 16, 160, 164ff, 193–194
CMYK space, 100 Conflation, 189–190
Cohen fundamental color space, 256 Contrast, 137, 331
Color appearance: Count of Rumford, 137, 159
models, 5 Crispening, 345
systems, 20–21, 358 chromatic, 229, 313ff, 318–319, 335
Color attributes, 6, 131ff, 134, 138ff, 342 lightness, 180, 313ff, 335, 345
Color circle, 46, 49ff, 55ff, 368 Cubo octahedron, 95, 295, 297
Color difference:
formula performance, 264ff, 347–348 Delboeuf, J. R. L., 66, 94
global, 332 Democritus, 22–22
just noticeable (JND), 11, 64–65, 83, Descartes, René, 15, 45
109, 202 De Valois model, 258–261
large, 238–239, 243, 266 Dietrich von Freiberg, 31
suprathreshold, 118, 225, 238–239, 266 DIN 6164 system, 93–95
threshold, 63–65, 83, 105, 112, 116, 216, DIN 99 formula, 244
266 DKL space, 174–175, 256, 258
unit contours, 322–325, 328–330 Donders, F., 78, 169
Color fundamentals, 161, 163 Doppler, C., 69
Color harmony, 45
Color Harmony Manual, 9, 89 Ebbinghaus, H., 66, 81, 177
Color matching error, 216, 266 Elementaristic approach, 2
data, 312–313 Ellipse fitting, 226–228
Color matching space, 351, 359 Empedocles, 21
Color mixing space, 100 Eraclius, 28
Color, object, 2, 7, 91–93 Euclidean geometry, 121, 190,
Color order systems, 19–21 348–349
Color, primary, 59, 69, 72, 368 Evans, R. M., 148
Color room, 47
Color solid, 1 Fechner, G. T., 64, 83, 105
psychological, 6, 121 Ficino, Marsilio, 33
psychophysical, 121 Field, George, 59
uniform, 352–353 Fludd, Robert, 37, 49
Color space, 1 Fluorence, 131
cone response, 164ff, 311, 337–338 Forsius, Sigfridus Aronius, 34–36, 49
cube root, 224 Friele FCM color difference formula,
ideal, 13, 332 232–234
 INDEX 405

Friele-MacAdam-Chickering formulas, scaling, psychophysical, 191–193

219–222 superimportance, 13, 151, 310
Frisch, J. C., 58 Hunt, R. W. G., 133–134
color appearance model, 269
Ganzfeld, 137, 340 Hunter color difference formula, 224
Gelb, A., 130 Hurvich & Jameson, 171–172, 179,
Geodesic, 205, 370 250–251
Gestalt psychology, 153–154, 370
Gerritsen, F., 91 Image framework, 154
Gilchrist, A. L., 138, 153–154 Indow, T., 120, 134, 143, 155, 203, 299
GLAB formula, 343 Intrinsic image model, 153
Glaukos (glaucus), 26, 31, 33, 34 Isomorphism, 121, 371
Glisson, Francis, 40–42
Godlove formula, 11–12 Jastrow, J., 83
Goethe, J. W. von, 53, 58, 63 Jameson & Hurvich, 171–172, 179,
Grassmann, H. G., 72, 125, 161 250–251
Gray scale, 40–42, 68, 118, 151–152, 176ff, Johannson, T., 91, 97, 303–304
277, 370 JPC color difference formula,
Greek color use, 25ff 234–235
Grossteste, Robert, 30 Judd, D. B., 11–13, 98, 149–151, 202,
Guild, J., 162 208–212, 248–249, 278, 286–287,
Guilford, J. P., 111 332
Guth ATD model, 253–255 Judd’s crinkled fan, 286
Judd’s value function, 178
Hardin, C. L., 102
Harris, Moses, 55–58 Kaneko, T., 179
Hayter, Charles, 71 Katz, D., 128, 153
Hefner lamp, 160 Kelly, K., 98
Helmholtz, H. L. F. von, 72–73, 123, Kircher, Athanasius, 39–40
125–126, 161, 205 Kirschmann, A., 80, 126
Helmholtz-Kohlrausch effect, 96, Klotz, Matthias, 58, 63
183–186, 292–293 König, A., 73, 160–161
Henry, Charles, 59 Kries, J. von, 78, 83, 105
Herbart, J. F., 64 Küppers, H., 91
Hering, K. E. K., 75–78, 96, 112, 127,
301–303 LABHNU formula, 234, 269
space, 9 Lambert, J. H., 53–55, 124–125
Hesselgren, S., 91, 97, 303–304 Land E., 19
Hickethier, E. A., 78, 91 Lateral geniculate nucleus 138, 371
Höfler, Alois, 78 Le Blon, J. C., 51
HSB color space, 100–101 Leonardo da Vinci, 32–33, 124
Hue, 351, 356 Leonardo 2000, 277
angle, 11–12, 100, 140–141, 145–146, Lichtenberg, G. C., 53, 363
150, 192, 195, 231, 242–247, Lightness, 34, 76, 78, 84, 107, 111,
326–332 124–125, 131–132, 151–154, 176,
coefficients, 141–142, 250 243, 355–356
difference, 325–328 heterochromatic 183, 341
unique, 4, 132, 138–140, 144–147, relative, 94
194–195, 340–342, 350 Limen, 64
406 INDEX

Line element, 204ff Optical Society of America

empirical, 216–218 Committee on Uniform Color Scales
visual space, 206 Subcommittee on Spacing of Munsell
Linking proposition, 131 Colors, 86
Lucretius, 27 Uniform color scales (OSA-UCS), 7,
Luminance, 161 13, 95–96, 255
Luo & Rigg data, 236, 312 atlas, 295–298
Luther-Nyberg color solid, 93, 169 formula optimization, 292–295
system development, 287–292
MacAdam, D. L., 91–92, 95, 117, 202, revised formula, 298–299
216–218, 222–223, 226, 287ff, uniformity, 299–300
298–299 Opponent color system, 261–263,
Macula, 169, 193–194, 371 350–351
Magnitude, 109 Opponent color theory, 78
Masking, 181, 372 Oresme, Nicole, 36
Maxwell, J. C., 72–74, 161 Ostwald, W., 9, 89–90, 96–97,
triangle, 208, 211 129
Mayer, Tobias, 51–53, 124
Measurement, 104 Paired comparison, 118
Metamers, 256, 264, 372 Palmer, George, 59, 363
Metaphorical imperative, 107 Photometry, 159
Middle Ages, 27 flicker, 160
Müller, A., 90 Perception, 2, 3, 15, 64, 113, 123–124
Müller, E. G., 128, 169, 176 empirically determined, 2, 102
Müller-Judd model, 248–249 Perceptibility, 228
Munsell, A. H., 9, 84–86, 128 Plateau, J. A. F., 66
1915 Atlas, 147, 177, 274 Plato, 22, 25
1929 Book of Color, 147, 177, Pliny, 25
275–277 Pope, A., 91
“color tree”, 9 Porphyry, 36
Renotations, 9, 86, 277, 281–283 Poussin, Nicholas, 51
Re-renotations, 189, 278 Power law, 110, 372
system development, 272–274 Pre-Platonists, 21
unit difference contour, 285 Principal component analysis, 101, 264,
Musschenbroek, P. van, 152, 364 342
Principal hue components, 155
Natural color system: Psychometric 133
Hering, 75–76, 127, 301 function, 119, 372
Swedish, 10, 96–98, 140, 182, 301ff Psychophysics, 6, 64ff, 104ff
atlas, 307–308 inner and outer, 65, 107
scale fitting, 304 Psychoquantitative, 133
terminology, 307–308 Ptolemy, 24–25
Neural correlate, 137 Pythagoras, 21
Neural network, 134
Newton, Isaac, 19, 43–45, 50, 124, 363 Reddish green, 263
Nickerson-Newhall color solid, 10 Reflectance, 5, 7, 16, 101–102, 134, 163,
Nickerson index of fading, 10, 149–150 359–360
Norwich’s law, 106 luminous, 153, 161, 176
 INDEX 407

Regression: Similitude, 137

linear, 226, 229 Smith-Pokorny functions, 166–167, 197,
nonlinear, 222 314
Retina, 4, 59, 76, 123, 160, 164, 373 Sowerby, James, 59
RGB space, 100 Spectral space, 101–102, 264, 359–360, 373
Rhombohedral, regular, 290, 292, 295, Speranskaya, N. I., 163
298 Stevens, S. S., 84, 111–113
Ridgway, R., 86 Stiles, W. S., 163, 206
color atlas, 86 Svaetichin, G., 171
Richter data, 216, 312 Surround, 2, 19–20, 180–181, 236, 243,
Riemann, B., 121 245, 309
geometry, 121, 148, 204, 288, 341, 346, SVF formula, 240–241
348, 373
plane, 328 Takasaki, H., 180
RIT-Du Pont data, 238, 268–269, 312 Telesio, Antonio, 33
Rods, 4, 160, 164 Tetractys, 36, 373
Rood, O. N., 84 Theophilus, 28ff, 124
Rösch-MacAdam limits, 89, 91, 121, 129, Theophrastus, 22, 24
164, 359 Threshold:
Rubens, Peter Paul, 37, 159 brightness, 152
Runge, Phillip Otto, 59–62, 124–125 chromatic, 148–149
Russell, B., 19 data, 237–238
Titchener, E. B., 66, 81–82, 202
Saturation, 50, 94–95, 100, 125–126, 128, Torgerson, W. S., 119
132, 373 Transformation:
Saunderson-Milner Zeta space, 186, 215 Linear, 162, 169, 194, 208ff, 230
Scales: nonlinear, 222–223
absolute, 116 projective, 208ff
category, 112, 118 Trichromatic, 4, 161, 164, 169, 373
chroma, 186–190 Tristimulus, 92, 163–164, 373
hue, 191–193
interval, 115 Unger, F. W., 59
lightness, 176ff Universal color language, 98–100, 136
minimally distinct border, 118 Urfarben (fundamental colors), 76, 302
multidimensional, 119–121, 134 Urso de Salerno, 30
nominal, 114
ordinal, 114 Value, 86, 128
ratio, 115 Veckenstedt, F., 26
tonal, 9, 27–31, 33, 34, 37, 39, 50, Vilialobos system, 91
57–58, 128 Vision:
Schiffermüller, I., 152, 179 theory of, 123–124
Schönfelder, S., 152, 179 Visual purple, 160
Schrödinger, E., 92, 205–206 Vollfarben, 10, 302, 368
Semmelroth. C. C., 189, 293 Vos-Wallraven line element, 207–208
Sensation, 15, 64–66, 84, 105–107
Shepard, R. N., 119, 138, 341 Wailer, Robert, 42–43
Signal detection theory, 106 Wavelength, dominant, 95, 131, 194–195,
Silberstein, L., 288, 290 305, 369
408 INDEX

Weber, E. H., 64 Wyszecki-Fielder data, 216, 312ff,

Weber-Fechner law, 78, 89, 109, 111–112, 333–334
127, 205 Wyszecki-Sanders data, 183, 255–256
Weber fraction, 110, 374
Weight integration, 270 Xenophanes, 21
Whiteness, 9, 10, 44, 76, 89, 109, 129–130
Witt data, 237, 239 Yellowish blue, 263
Wright, W. D., 162, 216–217 Young, Thomas, 59
Wundt, W. M., 82–83
Wyszecki, G., 14, 132, 211–212 Zone theory, 169–170, 369