See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/268813843

Color Coding for Data Visualization

Chapter · July 2014

DOI: 10.4018/978-1-4666-5888-2.ch161

CITATIONS READS
16 27,928

3 authors, including:

Simone Bianco Francesca Gasparini

Università degli Studi di Milano-Bicocca Università degli Studi di Milano-Bicocca
203 PUBLICATIONS 5,395 CITATIONS 135 PUBLICATIONS 1,935 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Simone Bianco on 12 January 2015.

The user has requested enhancement of the downloaded file.

Encyclopedia of
Information Science and
Technology, Third Edition

Mehdi Khosrow-Pour
Information Resources Management Association, USA

A volume in the
Category: Data Mining and Databases 85

Color Coding for Data Visualization D

Simone Bianco
Università degli Studi di Milano-Bicocca, Italy

Francesca Gasparini
Università degli Studi di Milano-Bicocca, Italy

Raimondo Schettini
Università degli Studi di Milano-Bicocca, Italy

INTRODUCTION BACKGROUND

Visualization is a process of mapping data onto visual A color coding scheme can be characterized by a color
dimensions to create a pictorial representation. A suc- mapping function f: D → C that maps data values D
cessful visualization provides a representation which to colors from the color palette C. In case of univari-
allows the user to gain insight into the structure of ate data, each value is mapped to a single color, with
the data or to communicate aspects of this structure multivariate data, each combination of values is mapped
effectively, The use of color for encoding information to a single color.
can greatly improve the observer’s understanding of the In the following we make a distinction only between
information depicted by image and his/her capacity for qualitative (nominal) and quantitative (ordinal, interval
remembering it. However, many aspects of color itself and ratio) data values.
and of its use are unknown to both users and system Associating a set of colors with a set of items to
designers. Users are often allowed to choose colors that express the significance of each is called “nominal color
can not be reproduced, that can not be distinguished by coding.” Examples of nominal data values are: water,
the human eye, or are easily misinterpreted. The broad vegetation, and urban. There is no logical ordering in
range of variables involved, as well as the interactions this sequence.
and the trade-offs among them, pose difficult problems Color can be also used in a quantitative fashion,
in selecting specific colors, in predicting their appear- i.e. to convey information about ordered data set. We
ance when they are seen in relation with others, and can distinguish here among “ordinal, interval and ratio
in predicting the observers’ interpretation and reaction color coding.” In ordinal coding the data values are in
to them. These problems can not be solved without a some way ordered, i.e., the data values can be put into
working knowledge about color reference systems, a sequence but no distance is defined between data
color reproduction technologies, mechanism of color values. Examples of ordinal data values are: very bad,
perception, users’ cultural and emotional reactions. bad, average, good, very good. In interval coding data
The goal of this paper is to critically discuss how a one can define a distance between two data values,
color scale should be designed to effectively represent but the zero point is arbitrary. The hue by itself can
both qualitative and quantitative information. Color be seen as interval data values, we can say that the
coding requirements, with regard to the task at hand, distance between red and yellow is 90° but we can not
the characteristics of the media, and the characteristics says that yellow is bigger than red. Periodic functions
of data to be coded are therefore analyzed providing typically produce interval data values. Data sets with
examples and references to related works. both positive and negative values can have a zero point
representing no change, average, or expected value. In
such data, deviation from zero is what is interesting.
In ratio coding a zero point is therefore defined. Data
values that represent the temperature are ratio data, in

DOI: 10.4018/978-1-4666-5888-2.ch161

Copyright © 2015, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Category:
Color Coding
Data
forMining
Data Visualization
and Databases

this case the zero point is defined at 0 Celsius degrees (http://www.vischeck.com/vischeck) is a web tool base
and we can state that a temperature of 20 Celsius on based on SCIELAB which shows how an image is D
degrees is twice a temperature of 10 Celsius degrees. seen by a user with some kind of color blindness. Apart
For nominal and ordinal data types discrete palettes from its aesthetic appeal, we can use image processing
must be used. For interval and ratio data one can use techniques to try to make information in images avail-
both discrete and continuous palettes, according to able to color blind people (http://www.vischeck.com/
the data structure and to the visualization aims. If a daltonize/) without distorting the color balance to an
discrete palette is used to code interval and ratio data, unacceptable degree. Color schemes that accommodate
those values must be a-priori quantized into a finite red or green-blind dichromats will accommodate most
number of ranges. other forms of color deficiency.
Many colors have certain qualities associated with
them because of their natural occurrence, cultural usage
MAIN FOCUS OF THE ARTICLE or technical norms. However color also depends on,
and is constrained by, the cultural traditions or techni-
The goal of this paper is to critically discuss how a color cal experience of the users. We commonly associate
scale should be designed to be effectively adopted to green with vegetation, for an USA audience the color
obtain a successful visualization. For instance, adding green is associated with the color of money.
a color which does not add additional insight to the Thus in this paper the color coding requirements,
visualization can sometimes cause confusion as users with regard to the task at hand, the characteristics of
try to understand its meaning and should, therefore, the media, and the characteristics of data to be coded
be avoided. So, it is particularly important to perform are therefore analyzed providing examples and refer-
the right choice in order to build visualizations which ences to related works.
depict the desired information in a clear way. The color
scales and methods used for encoding information Color Space Selection
depend on several factors. They certainly depend on
the data type involved in terms of information carried Several approaches to color description, whatever
(qualitative and quantitative) and dimensionality of based on device-dependent data, colorimetry, or some
the representations (univariate or multivariate). The other empirical organizations give rise to various color
goal of a specific visualization is also an important spaces. The use of RGB color spaces seems unrealistic
issue when choosing a color scale. Different tasks and for man-machine communications since its metrics
goals require different color coding schemes. Certain does not represent color differences in a uniform scale
goals may require specific subsets of data to be high- and the colors are not organized in a manner intuitive
lighted. This can be achieved for instance by using for human observers. Color spaces, such as RGB and
bright, warm, and fully saturated colors. Within this derived color spaces, are device-dependent and there-
context, Bergman et al. (1995) have proposed to select fore do not univocally describe colors. If colors were
the appropriate color scales according to four specific identified only by their device-dependent coordinates
tasks, while Tominski et al. (2008) have presented a it might be difficult to reproduce the same image on
task-driven color coding. Different color perception of another device (display, printer, projector….) or worst
the viewers must also be considered. An estimated nine still, an image that differs from the one intended is
to twelve percent of the male population and less than generated.
1% of females suffer from some form of color vision Device-independent color description, perceptual
deficiency. The degree to which a person may have an addressability, and the capability of representing color
abnormal color vision ranges from slight difficulty in differences in a uniform scale can be achieved by speci-
recognizing color shades to a complete loss of color fying colors in the CIELUV or CIELAB color spaces.
vision. The most prevalent type is a deficiency in per- In our opinion CIELAB space represents the best that
ceiving red/green differences (approximately eight to can be achieved with a space of Euclidian metrics.
ten percent of the male population), while one to two However, as it does not take into account all factors
percent of men are blue/yellow color blind. VisCheck

86
 Category:
Color Coding
Data Mining
for Data
andVisualization
Databases

determining color response, it should not be used as Color Management Systems, interested readers may
an absolute standard for color appearance description. refer to http://www.color.org/index.xalter. D
Obviously, other color spaces can be used for select-
ing colors in particular for nominal color coding. One Qualitative Color Coding
could, for example, define a color atlas by sampling the
color solid obtained by mapping the RGB color cube Color is pre-attentively observed and used to segment
of the device to the CIELAB color space. Campadelli the visual environments into objects. This characteristic
et. al (2001) have preferred to use the Munsell Atlas makes it particularly effective in coding qualitative
for nominal color coding mainly for two reasons: information. In nominal color coding all colors should
be distinguishable, the palette should display no clear
• The classified images may not only be displayed ordering and all colors should be perceived as equally
on the screen but also printed. Since the gamut important. There should be no perceptual ordering
of feasible colors of common displays and col- in the representation. The number of colors used to
or printers differ considerably, a “good,” high represent nominal data should be restricted to seven or
contrast, color coding on the screen may be less. The problem is that we have often more items to
drastically modified when printed. Many colors represent than easily discriminable colors. Moreover
in the design of a coding, colors may be used not only
(in the ranges of blues and greens in particu-
to permit the recognition or discrimination of informa-
lar) cannot be reproduced in print. Restricting
tion items, but also to convey high-level information
the selection of color to the Munsell color atlas
at local (e.g. areas are evaluated) and/or global level
greatly limits this drawback. (e.g. apparent conceptual linking between different
• Munsell specification can be associated with items) as well, (Wang et al. 2008).
the ISCC-NBS standard color naming system Several heuristic procedures have been proposed
providing a simple way of verifying whether to define high-contrast sets of colors. Kelly (1976) has
the colors selected, either manually or auto- conceived a list of 22 maximally contrasting surface
matically, have similar linguistic descriptions. colors, such that each color of the list is maximally
This is very helpful in preventing undesirable different from the one immediately preceding it. In
conceptual linking between unrelated classes in 1982, Carter and Carter (1982) formulated the first
the classified image, or, viceversa, in creating algorithm to compute easily discriminable sets of
those links when necessary. colors: selecting a subset of highly contrasting colors
from a given range of colors means choosing a subset
such that in it the minimal distance among all possible
To map the colors in RGB values for reproduction
couples of colors is maximal. Campadelli et al. (1995)
on a given output device, the user must provide the
have presented an abstract formulation of the problem
color profile of the display device, a well-defined tool
of selecting high-contrast color sets, defining it as a
enabling color conversion in cross-media color repro-
combinatorial optimization problem on graphs. They
duction. The International Color Consortium (ICC)
also addressed the algorithmic aspect of the selection
has defined the rules for coding in ICC profiles, the
of high-contrast color sets. They presented an effec-
functions for the conversion of colors from a device-
tive algorithm, which has an innovative feature that
independent to a device-dependent color space, together
could also be useful when one has to take into account
with information about the viewing conditions and
visual features such as the shape, size, and texture of
media characteristics recommended for Color Manage-
the graphical items to be coded. The algorithm does
ment Systems, which can be used to ensure faithful
not require that the function used to code the similarity
color reproduction on different media or devices. If
between two colors (or graphical codes obtained by
the user cannot supply the ICC profile, a default dis-
combining color with other visual features) be a dis-
play device set to the Standard RGB color space and
tance function; in other words, it works whatever the
corresponding color profile should be assumed. For a
measure of perceptual similarity selected (Campadelli,
more complete discussion of color reproduction and
2001). Several authors have devised algorithms that

87
Category:
Color Coding
Data
forMining
Data Visualization
and Databases

Figure 1. The sets of colors (with cardinalities from 2 to 25) selected applying the method described in Bianco
(2013), rendered in sRGB under the CIE D65 illuminant D

can also fulfill a number of ergonomical requirements browsing the Munsell color space and setting percep-
(Bergman, 1995; Borland, 2011; Della Vantura, 1991). tual constraints on the colors the system then selects
Bianco and Citrolo (2013) addressed the algorithmic automatically. This system is available for academic
aspect of the selection of high contrast color sets under and non-profit purposes. ColorBrewer (is an online tool
single or multiple illuminants, and present an effective designed to help people select good color schemes for
algorithm based on Local Search, which is fast enough maps and other graphics. The system informs the user
to be used in real time applications. The sets are ex- about several important properties of the chosen color
tracted from the Munsell atlas and have cardinalities representation, such as if it is suitable for color blind
from 2 to 25. The method is also generalized to cope people and if can be faithfully color printed. It is free
with multiple illuminants simultaneously (Figure 1). to use. In Figure 2, the interface layouts of Quickcolor
Borland and Huber (2011) argued that designing and ColorBrewer are shown.
color maps for domain experts differs fundamentally When the user defines the color coding, setting
from designing them for general audiences. Domain- the colors selected in correspondence with the values
specific knowledge should inform the design choices of the data, some simple guidelines should be taken
and domain experts need to understand the design into account (Della Ventura, 1993, Lefkowitz, 1992):
choices, a close collaboration between designers and
experts is therefore necessary. • The human eye is only effective in detecting
There are some systems aimed to facilitate this co- relative values of colored areas in close spatial
operation. Campadelli et al. (2001) presented a system proximity. If these are far apart, local distor-
for the automatic selection of conspicuous color sets tions may somewhat alter the information.
for qualitative data display called Quickcolor. Based
• The edges formed by chromatic differences
on visual interaction, it provides effective tools for
alone with no lightness difference are poor

88
 Category:
Color Coding
Data Mining
for Data
andVisualization
Databases

Figure 2. Quickcolor (on the left), and ColorBrewer (on the right) interface layouts for nominal color coding
D

guides to accurate focusing: these edges remain that certain elements differ in their properties from the
fuzzy. others. However we can also use color for association,
• Colors that differ only in the amount of blue do that is, for indicating that certain elements in a design
not produce a sharp edge. have common properties. Nominal color coding is
widely employed by the image processing community
• The eye can not maintain its focus various
to represent the output of a classification-segmentation
on highly-saturated highly-bright objects in
process. In this context a serious problem arises because
spectrally extreme hues (e.g. blue-red, yellow-
the chosen colors must be displayed together and as-
purple) at the same time (a 3D illusion is pro- signed to classes composed of regions of different size
duced). Desaturating these colors reduces the and morphology. The user must take into account the
need for refocusing. characteristics of the image (the number of classes, the
• The color of a patch affects its perceived area. links between them, and the geometric and topological
Saturated colors appear larger than desaturat- features that belong to the different classes) so that the
ed ones. It has been shown that when highly association of classes with colors produces a readable,
saturated red and green were used to color a pleasant coded image. Although various systems have
statistical map, the red areas were judged to be been proposed to support the user in color coding,
larger, even though the regions were actually these are mainly based on a trial-and error approach
of the same size. No such consistent distortion and demand a concerted effort on the part of the user
(Campadelli, 1994).
occurred when low saturated colors were used.
• Since the smaller the area, the less discrim-
inable the colors, the user should be encour- Quantitative Color Coding
aged to limit the use of highly saturated and
When we want to code a quantitative information, we
bright colors to rendering perceivable thin lines
must compromise between the need for correct per-
and small shapes in the image.
ception of a single color (so that it can be translated,
• Areas of strong color can produce afterimages using the legend, into a numerical value) and the need
when the viewer looks away from the screen. to convey to the viewer the order intrinsic to the data.
Bright, highly saturated colors should not used In some cases convention determines the choice of a
for prolonged viewing but just to grab attention. color scale and the assignment of colors to data values,
for example geographers have a well-defined scale to
In the framework of nominal color coding, we usu- display height above sea level. In many instances there is
ally use color for differentiation, that is for indicating no conventional reason for choosing one color sequence

89
Category:
Color Coding
Data
forMining
Data Visualization
and Databases

over another, as is the case for a map that illustrates surface shape, while saturation and hue suitable for
the earth’s magnetic field. Trumbo’s principles seem convey low spatial frequency information in the coded D
important in generating and evaluating an ordinal color data. Color scales which monotonically increase in
scale (Trumbo, 1981): luminance are good candidates for representing data
with a high spatial frequency. A color scale without
• Principle I (Order): If the data are ordered, luminance variation will not be able to convey fine
then the colors chosen to represent them should spatial information, while a color scale which only
be perceived as preserving that order. varies in luminance may effectively code large-scale
structural composition and variation but may not be
• Principle II (Separation): Important differ-
able to depict information about low frequency, gradual
ences in data values should be represented by
changes. A color scale that varies in both luminance
colors clearly perceived as different.
and hue can be used to accurately represent both metric
and surface properties.
To the above principles, Brewers add a third re- The use of RGB color spaces is not appropriate
quirement for cartography (http://colorbrewer2.org/): for quantitative color coding since its metrics does not
the colors should be perceived as equally important. represent color differences in a uniform scale and the
Grey scale/luminance/brightness and saturation scales color are not organized in an intuitive manner for hu-
have an innate visual order. While, disregard less of its man observers. Color is described more meaningfully
popularity rainbow scales are not particularly useful in terms of the perceptual dimensions of lightness,
in conveying a sense of order in the data, since they hue, and colorfulness than in terms of device signals.
have no intuitive beginning or end. Heated-object scale Transformations have been developed to map the RGB
represents a compromise between the luminance and color space into more intuitive ones such as HSV, HIS,
the rainbow scale. It goes from black to white, passing HLS. However, the perceptual dimensions defined
through dark red, to orange and to light yellow. This within these color spaces are only a rough approxima-
color scale has a perceived natural ordering since tion of the psychological dimensions. For interval and
it monotonically increases in brightness. Lightness ratio color coding, equal steps in data values should
scales, color scales that increase monotonically in correspond to equally perceived magnitude in the
lightness and saturation, or color scales that increase representation. A perceptually uniform color space
monotonically in lightness while cycling through: i.e. can be used to help choosing the appropriate colors.
range of hues (short sections of the hue circle may be Bivariate color scales provide a method for encoding
perceived as a ”continua” e.g. red, orange and yellow) two data sets into one producing a visual representation
are more successful in conveying a sense of order. This that allows the observer to interpret these two sources
redundant representation also has the advantage that it of information simultaneously. The simplest bivariate
can be unambiguously interpreted by someone with a schemes use planes or surface which are constant along
dichromatic color deficiency. It has been empirically a perceptual variable (hue, chroma, or lightness). A
demonstrated that when a reference scale (legend) is number of bivariate color palettes using various com-
shown, color scales including hue variations are much binations of hue and lightness color scales have been
to be preferred (in terms of readability) to others. With proposed and investigated (Ware, 1988). Some bivariate
no legend, luminance encoding is far worse than the and trivariate color scales can be hard to understand
other coding. for an untrained user (Pham, 1990).
Ware (1988) classified the information extracted As examples of different color scales, we here
from color coded maps into metric and form. Metric consider the built-in colormaps supported by MAT-
information conveys quantitative information at each LAB (www.mathworks.com). The built-in colormaps
point; form information conveys the shape or structure (in the case of 128 colors) are shown in Figure 3 and
of the data. The human eye has different Contrast Sen- described below.
sitivity Functions (CSF) for luminance and chromatic
channels. These differences imply that luminance
• Colorcube: Contains as many regularly spaced
is more suitable for carrying high spatial frequency
colors in RGB color space as possible, while
information and plays and depict image structure and

90
 Category:
Color Coding
Data Mining
for Data
andVisualization
Databases

Figure 3. The MATLAB built-in colormaps

D

attempting to provide more steps of gray, pure • Bone: Is a grayscale colormap with a higher
red, pure green, and pure blue value for the blue component.
• Hsv: Varies the hue component of the hue- • Copper: Varies smoothly from black to bright
saturation-value color model. The colors be- copper.
gin with red, pass through yellow, green, cyan, • Pink: Contains pastel shades of pink. The pink
blue, magenta, and return to red. This colormap colormap provides sepia tone colorization of
is particularly appropriate for displaying peri- grayscale photographs.
odic functions. • Flag: Consists of the colors red, white, blue,
• Jet: Ranges from blue to red, and passes and black. This colormap completely changes
through the colors cyan, yellow, and orange. color with each index increment.
It is a variation of the hsv colormap. It corre- • Prism: Repeats the six colors red, orange, yel-
sponds to the rainbow colormap. low, green, blue, and violet.
• Hot: Varies smoothly from black through
shades of red, orange, and yellow, to white. In Figures 4, 5, 6, and 7 some example images with
• Cool: Consists of colors that are shades of cyan different contents are visualized using different color
and magenta. It varies smoothly from cyan to scales. The reader may note how the data forms and
magenta. values are differently coded and thus perceivable. It
could also be noted how different color scales make
• Spring: Consists of colors that are shades of
more or less visible high frequency contents.
magenta and yellow.
• Summer: Consists of colors that are shades of
green and yellow.
FUTURE RESEARCH DIRECTIONS
• Autumn: Varies smoothly from red, through
orange, to yellow.
The relationship between color combinations that have
• Winter: Consists of colors that are shades of
harmony, which is defined as a set of colors that are
blue and green. aesthetically pleasing together to the human perceptual
• Gray: Returns a linear grayscale colormap. system, and visualization effectiveness is as topic that
are gathering attention in the in the visualization com-

91
Category:
Color Coding
Data
forMining
Data Visualization
and Databases

Figure 4. Different MATLAB colormaps applied to a synthetic image

D

munity in and reserves further research (Einakian & colors by collecting representative images, analyzing
Newman 2010). image color distributions to determine value-color
Concepts can invoke colors, and vice versa. In Lin et affinity scores, and choosing an optimal assignment.
al. (2013) the authors investigate concept-color associa- Annotation of an image with text is a commonly
tions to design effective categorical color assignments encountered task. Garg et al. present a prototype sys-
for visualization. Such mappings may allow people to tem that helps users by suggesting appropriate colors
use more automated pathways to process value-color for inserting text and symbols into an image. The
associations, require less conscious thought. Practically, color distribution in the image regions surrounding
concept-color associations improved recognition of the annotation area determines the colors that make a
category values may reduce the need to consult the data good choice – i.e. capture a viewer’s attention, while
legend. Given a set of categorical values and a target remaining legible. Future work includes incorporating
color palette, the proposed algorithm matches each aesthetics giving scores to colors based on color har-
data value with a unique color. Values are mapped to

Figure 5. Different MATLAB colormaps applied to a portrait image having regions of different spatial frequency

92
 Category:
Color Coding
Data Mining
for Data
andVisualization
Databases

Figure 6. Different MATLAB colormaps applied to a sample image

D

mony and extending the system to highlight elements REFERENCES

volume datasets.
Bergman, L. D., Rogowitz, B. E., & Treinish, L.
A. (1995). A rule-based tool for assisting colormap
CONCLUSION selection. In Proceedings of the 6th conference on
Visualization’95 (p. 118). IEEE Computer Society.
Color is a basic visual feature, which, together with Bianco, S., & Citrolo, A. G. (2013). High contrast color
shape, size, and texture and a few other features, char- sets under multiple illuminants. In Computational Color
acterizes an image. Color is pre-attentively observed, Imaging (pp. 133–142). Springer Berlin Heidelberg.
this characteristic makes it particularly effective in doi:10.1007/978-3-642-36700-7_11
conveying qualitative and quantitative information in
images, maps, graphics illustrations, and diagrams. The Borland, D., & Huber, A. (2011). Collaboration-
goal of this paper has been to critically discuss how a Specific Color-Map Design. Computer Graphics
color scale should be designed to effectively represent and Applications, IEEE, 31(4), 7–11. doi:10.1109/
both qualitative and quantitative information. We have MCG.2011.55 PMID:24808154
discussed color coding requirements with regard to the Campadelli, P., Mora, P., & Schettini, R. (1994, Oc-
task at hand, the characteristics of the media, and the tober). Using Hopfield networks in the nominal color
characteristics of data to be coded. We have shown that coding of classified images. In Pattern Recognition,
the choice of the proper color representation scheme 1994. Vol. 2-Conference B: Computer Vision & Image
to use with a particular data and communication task Processing., Proceedings of the 12th IAPR Interna-
is not just an esthetic matter. Nonetheless, taking into tional. Conference on (Vol. 2, pp. 112-116). IEEE.
account the fundamental principles of human vision
and the characteristics of the output device, an effec- Campadelli, P., Mora, P., & Schettini, R. (1995). Color
tive color coding is possible providing representations set selection for nominal coding by Hopfield networks.
which communicate aspects of both the data values The Visual Computer, 11(3), 150–155. doi:10.1007/
and the structure effectively. BF01898600

Figure 7. Different MATLAB colormaps applied to render the shapes of 3D objects

93
Category:
Color Coding
Data
forMining
Data Visualization
and Databases

Campadelli, P., Schettini, R., & Zuffi, S. (2001). A Wang, L., Giesen, J., McDonnell, K. T., Zolliker, P.,
system for the automatic selection of conspicuous color & Mueller, K. (2008). Color design for illustrative D
sets for qualitative data display. Geoscience and Remote visualization. Visualization and Computer Graphics.
Sensing. IEEE Transactions on, 39(10), 2283–2286. IEEE Transactions on, 14(6), 1739–1754.

Carter, R. C., & Carter, E. C. (1982). High-contrast Ware, C. (1988). Color sequences for univariate
sets of colors. Applied Optics, 21(16), 2936–2939. maps: Theory, experiments and principles. Com-
doi:10.1364/AO.21.002936 PMID:20396153 puter Graphics and Applications, IEEE, 8(5), 41–49.
doi:10.1109/38.7760
Della Ventura, A., & Schettini, R. (1992). Computer-
aided color coding for data display. In Pattern Recogni-
tion, 1992. Vol. III. Conference C: Image, Speech and
Signal Analysis, Proceedings., 11th IAPR International ADDITIONAL READING
Conference on (pp. 29-32). IEEE.
Bertin, J. (1983). Semiology of graphics: Diagrams,
Einakian, S., & Newman, T. S. (2010, January). Experi-
networks, maps (W. J. Berg, Trans.). Madison, WI:
ments on effective color combinations in map-based
The University of Wisconsin Press, Ltd.
information visualization. In IS&T/SPIE Electronic
Imaging (pp. 753005-753005). International Society Brodlie, K., Carpenter, L., Earnshaw, R., Gallop, J.,
for Optics and Photonics. Hubbold, R., & Mumford, A. et al. (1992). Scientific
visualization, techniques and applications. New York:
Garg, S., Padalkar, K., & Mueller, K. (2011). Magic
Springer-Verlag. doi:10.1007/978-3-642-76942-9
marker: a color analytics interface for image annota-
tion. In Advances in Visual Computing (pp. 629–640). Campadelli, P., Mora, P., & Schettini, R. (1994).
Springer Berlin Heidelberg. doi:10.1007/978-3-642- Nominal Color Coding of Classified Images by Hop-
24028-7_58 field Networks. In ICANN’94 (pp. 373-376). Springer
London.
Kelly, K. L. (1965). Twenty-two colors of maximum
contrast. Color Engineering, 3(26), 26–27. Campadelli, P., Posenato, R., & Schettini, R.
(1999). An algorithm for the selection of high-
Levkowitz, H., & Herman, G. T. (1992). The design
contrast color sets. Color Research and Applica-
and evaluation of color scales for image data. IEEE
tion, 24(2), 132–138. doi:10.1002/(SICI)1520-
Computer Graphics and Applications, 12(1), 72–80.
6378(199904)24:2<132::AID-COL8>3.0.CO;2-B
doi:10.1109/38.135886
Campadelli, P., Schettini, R., & Zuffi, S. (2000). An
Lin, S., Fortuna, J., Kulkarni, C., Stone, M., & Heer,
Intelligent System for the Selection of Conspicuous
J. (2013). Selecting Semantically-Resonant Colors for
Color Sets. Journal of Image and Graphics, 5(Suppl.),
Data Visualization. Computer Graphics Forum (Proc.
500–503.
EuroVis), 32(3).
Carter, E. C., & Carter, R. C. (1981). Color and con-
Pham, B. (1990). Spline-based color sequences for
spicuousness. JOSA, 71(6), 723–729. doi:10.1364/
univariate, bivariate and trivariate mapping. In Pro-
JOSA.71.000723 PMID:7252614
ceedings of the 1st conference on Visualization’90 (pp.
202-208). IEEE Computer Society Press. Della Ventura, A., & Schettini, R. (1993). Computer
aided color coding. In Communicating with virtual
Tominski, C., Fuchs, G., & Schumann, H. (2008).
worlds (pp. 62–75). Springer Japan. doi:10.1007/978-
Task-driven color coding. In Information Visualisa-
4-431-68456-5_6
tion, 2008. IV’08. 12th International Conference (pp.
373-380). IEEE. MacDonald, L. W. (1999). Using color effectively in
computer graphics. Computer Graphics and Applica-
Trumbo, B. E. (1981). A theory for coloring bivari-
tions, IEEE, 19(4), 20–35. doi:10.1109/38.773961
ate statistical maps. The American Statistician, 35(4),
220–226.

94
 Category:
Color Coding
Data Mining
for Data
andVisualization
Databases

Robertson, P. K., & O’Callaghan, J. F. (1986). The KEY TERMS AND DEFINITIONS
generation of color sequences for univariate and bivari- D
ate mapping. Computer Graphics and Applications, Color Space: Mathematical model describing the
IEEE, 6(2), 24–32. doi:10.1109/MCG.1986.276688 way colors can be represented as tuples of numbers.
CMS: Color Management Systems.
Rogowitz, B. E., & Kalvin, A. D. (2001, October).
HVS: Human Visual System.
The which blair project: A quick visual method for
Qualitative Color Coding: Process of associat-
evaluating perceptual color maps. In Proceedings of
ing a set of colors with a set of items to express their
the conference on Visualization’01 (pp. 183-190). IEEE
significance.
Computer Society.
Quantitative Color Coding: Process of associating
Rogowitz, B. E., & Treinish, L. (2009). Why Should a color scale to represent information about ordered
Engineers and Scientists Be Worried About Color? data set.
IBM Thomas J. Yorktown Heights, NY: Watson Re- Visualization: Process of mapping data onto visual
search Center. dimensions to create a pictorial representation.

Rogowitz, B. E., Treinish, L. A., & Bryson, S. (1996).

How not to lie with visualization. Computers in Phys-
ics, 10(3), 268–273. doi:10.1063/1.4822401

Silva, S., Madeira, J., & Santos, B. S. (2007, July).

There is more to color scales than meets the eye: A
review on the use of color in visualization. In Proceed-
ings of the Information Visualization, 2007. IV’07.
11th International Conference (pp. 943-950). IEEE.

Tufte, E. R., & Graves-Morris, P. R. (1983). The visual

display of quantitative information (Vol. 2). Cheshire,
CT: Graphics press.

95

View publication stats