True Color: Component

Ordering
 Colors in 3 separate arrays of similar length
 Retrieve same location (u,v) in each R, G and B
array
True Color: Packed
Ordering
Component (RGB) values representing a pixel’s color
is packed together into single element
 Indexed
 Images
Permit only limited number of distinct colors (N = 2 to
256)
 Used in illustrations or graphics containing large regions
with same color
 Instead of intensity values, image contains indices into
color table or palette
 Palette saved as part of image
 Converting from true color to indexed color
requires quantization
Color Images in
ImageJ
 ImageJ supports 2 types of color images
 RGB full‐color images (24‐bit “RGB color”),
 packed order
 Supports TIFF, BMP, JPEG, PNG and RAW file formats
 Indexed images (“8‐bit color”)
 Up to 256 colors max (8 bits)
 Supports GIF, PNG, BMP and TIFF (uncompressed) file
formats
 See section 12.1.2 of Burger & Burge
Color Image Conversion in
ImageJ
 Methods for converting between different types of color
and grayscale image objects
 Note: if doScaling is true, pixel values scaled to
maximum range of new image
General Strategies
for Processing Color
Images
 Strategy 1:
Process each RGB
matrix separately
General Strategies
for Processing Color
Images
 Strategy 2: Compute
luminance
(weighted average
of RGB), process
intensity matrix
RGB to HSV
Conversion
 Define the following
values

 Find saturation of RGB color components (Cmax=

255)

 And luminance
value
RGB to HSV
Conversion
 Normalize each component
using

 Calculate preliminary hue value H’

as

 Finally, obtain final hue value by normalizing to interval

[0,1]
Example: RGB to HSV
Conversion

Original RGB
image

HSV values
in grayscale
HSV to RGB
Conversion

HSV to RGB
Conversion
RGB Components can be scaled to whole numbers in
range [0,255] as
RGB to HLS
Conversion
 Compute Hue same way as for HSV
model

 Then compute the other 2 values

as:
Example RGB to HLS
Conversion
Original RGB
image

HLS values
in grayscale
HLS to RGB
Conversion
 Assuming H, L and S in [0,1]
range

 Otherwise,
calculate

 Then calculate the

values
HLS to RGB
Conversion
Assignment of RGB values is done as
follows
TV Color Spaces – YUV, YIQ,
YC b C r
 YUV, YIQ: color encoding for analog NTSC and PAL
 YCbCr: Digital TV encoding
 Key common ideas:
 Separate luminance component Y from 2
chroma components
 Instead of encoding colors, encode color differences
between components (maintains compatibility with
black and white TV)
YU
V
 Basis for color encoding in analog TV in north america
(NTSC) and Europe (PAL)
 Y components computed from RGB components as

 UV components computed
as:
YU
V
 Entire transformation from RGB to
YUV

 Invert matrix above to transform from YUV back to

RGB
YI
Q
 Original NTSC used variant of YUV called YIQ
 Y component is same as in YUV
 Both U and V color vectors rotated and mirrored so
that

2D rotation
matrix

where β = 0.576 (33

degrees)
YCbC
r Internationally standardized variant of YUV


 Used for digital TV and image compression (e.g.

JPEG)
 Y, Cb, Cr components calculated as

 Inverse transform from YCbCr to RGB

YCbC
r
 ITU recommendation BT.601 specifies values:
wR = 0.299, wB = 0.114, wG = 1 – wB – wR =
0.587
 Thus the transformation

 And the inverse transformation

becomes
Comparing YUV, YIQ and
YCbCr
 Y values are identical
in all 3 color spaces
CIE Color
Space
 The gamut of all colors perceivable is thus
a three‐dimensional shape in X,Y,Z
 Color = X’X + Y’Y + Z’Z
CIE Chromaticity Diagram
(1931)
•For simplicity, we often
project to the 2D plane
•Also normalize

•Note: Inside horseshoe visible,

outside invisible to eye

Note: Look up x, y
Calculate z as 1 – x - y
Standard
Illuminants
 Central goal of CIE chart is the quantitative measurement
of colors in physical reality
 CIE specifies some standard illuminants for many real and
hypothetical light sources
 Specified by spectral radiant power distribution
and correlated color temperature
 D50: natural direct sunlight
 D65: Indirect daylight, overcast sky
 CIE
uses
Find complementary colors:
 equal linear distances from white in opposite directions
 Measure hue and saturation:
 Extend line from color to white till it cuts horseshoe (hue)
 Saturation is ratio of distances color‐to‐white/hue‐to‐white
 Define and compare device color gamut (color
ranges)
 Problem: not perceptually uniform:
 Same amount of changes in different directions
generate perceived difference that are not equal
 CIE LUV, L*a*b* ‐ uniform
CIE
L*a*b*
 Main goal was to make changes in this space linear
with respect to human perception
 Now popular in high‐quality photographic applications
 Used in Adobe photoshop as standard for converting
between different color spaces
 Components:
 Luminosity L
 Color components a* and b* which specify color hue and
saturation along green‐red and blue‐yellow axes
 All 3 components are measured relative to reference
white
 Non‐linear correction function (like gamma correction)
applied
Device Color
Gamuts
 Since X, Y, and Z are hypothetical light sources, no real
device can produce the entire gamut of perceivable
color
 Depends on physical means of producing color on
device
 Example: R,G,B phosphors on CRT monitor
 Device Color
Gamuts
 The RGB color cube sits within CIE color space
 We can use the CIE chromaticity diagram to compare
the gamuts of various devices
 E.g. compare color printer and monitor color gamuts
Transformation CIE XYZ to
L*a*b*
 Current ISO Standard 13655
conversion

where

 Usually, standard illuminant D65 is specified as

reference (Xref, Yref, Zref)

 Possible values of a* and b* are in range [‐127, +127]
Transformation L*a*b* to CIE
XYZ
 Reverse transformation from L*a*b* space to XYZ
is

where
Example of L*a*b*
Components
Code for XYZ to L*a*b*
and L*a*b* to XYZ
Conversion
Measuring Color
Differences
Due to its uniformity with respect to human
perception, differences between colors in L*a*b*
color space can be determined as euclidean distance
sRG
B
 For many computer display‐oriented applications, use of
CIE color space may be too cumbersome
 sRGB
 developed by Hewlett Packard and Microsoft
 has relatively small gamut compared to L*a*b*
 Its colors can be reproduced by most computer monitors
 De Facto standard for digital cameras
 Several image formats (EXIF, PNG) based on sRGB
Transformation CIE XYZ to
sRGB
 First compute linear RGB values
as

Where

 Next gamma correct linear RGB

values
Transformation sRGB to CIE
XYZ
 First linearize R’ G’ B’ values
as

With

 Linearized RGB then transformed to XYZ

as
Adobe
RGB
 Small gamut limited to colors reproducible by
computer monitors is a weakness of sRGB
 Creates problems in areas such as printing
 Adobe RGB similar to sRGB but with much larger
gamut

sRGB gamut Adobe RGB gamut

Chromatic
Adaptation
 Human eye adapts to make color of object same
under different lighting conditions
 E.g. Paper appears white in bright daylight and
under flourescent light
 CIE color system allows colors to be specified relative to
white point, called relative colorimetry
 If 2 colors specified relative to different white points, they
can be related to each other using chromatic adaptation
transformation (CAT)
XYZ
Scaling
 Simplest chromatic adaptation method is XYZ
scaling
 Basically, color coordinates multiplied by ratios
of corresponding white point coordinates

 For example to convert colors from system based on

white point W1 = D65 to system relative to W2 = D50

 Another alternative is Bradford

adaptation
Colorimetric Support in
Java
 sRGB is standard color space in Java
 Components of color objects and RGB color images are
color corrected
Statistics of Color
Images
 Task: Determine how many unique colors in a given image
 Approach 1: Create histogram, count frequency of each
color
 Not efficient since for 24‐bit image, there are 224 = 16,777,216
colors
 Approach 2: Put pixels in array, then sort
 Similar colors next to each other
 We can then count the number of transitions between
neighboring colors
 Much more efficient than histogram approach if many repeated
colors
Color
Histograms
 When applied to object of type ColorProcessor, built‐in
ImageJ method getHistogram( ), simply counts intensity
of corresponding gray values
 Another alternative: compute individual intensity
histograms of 3 color channels
 Downside: does not give information about actual colors in image
 A good alternative: 2D color histograms
 2D Color
Histograms
Define:

 Note: * denotes arbitrary component value

 Resulting 2D histograms are independent of original
image size, easily visualized
Example:
Combined Color
Histogram
Color
Quantization
 True color can be quite large in actual description

 Sometimes need to reduce size

 Examples:
 Convert 24‐bit TIFF to 8‐bit TIFF
 Take a true‐color description from database and convert to
web image format
 Replace true‐color with “best match” from smaller
subset
 Quantization algorithms:
 Uniform quantization
 Popularity algorithm
 Median‐cut algorithm
 Octree algorithm