Chapter 3
Color Feature Extraction
Color is an important and the most straight-forward feature that humans
perceive when viewing an image. Human vision system is more sensitive
to color information than gray levels so color is the first candidate used
for feature extraction. Color histogram is one common method used to
represent color contents. The algorithms follow a similar procession: Se-
lection of a color space, representation of color features, and matching
algorithms.
3.1 Color Spaces Selection
There are several color spaces used internationally. Each of those color
spaces were designed for certain applications and based on certain con-
siderations.
3.1.1 RGB color space
RGB color space is the most common one used for images on computer
because the computer display is using the combination of the primary
colors (red, green, blue) to display any perceived color. Each pixel in the
screen is composed of three points which is stimulated by red, green and
blue electron gun separately. However, RGB space is not perceptually
uniform so the color distance in RGB color space does not correspond to
color dissimilarity in perception. Therefore we prefer to transform image
data in RGB color space to other perceptual uniform space before feature
extraction.
50 Chapter 3
3.1.2 Munsell color system
The Munsell color system organizes the colors according to natural at-
tributes. It represents three color primitives (hue, intensity and satura-
tion) by a sphere like 3 dimensional spatial model and names them as hue,
value and chroma. Its advantage is that it is a discrete system and gives a
finite set of colors by perceptual similarities over an intuitive three dimen-
sional space. Because of its perceptual uniformity, Munsell color system
is used widely for color design. The disadvantage is that there does not
exist a simple transformation from RGB color space and often requires a
lookup table to convert.
3.1.3 CIE color systems
The CIE (Commission Internationale de I' Eclairage) in 1976 defined
the CIE 1976 (L*u*v*) and CIE 1976 (L*a*b*) two color spaces for the
perceptual uniformity. They are produced by linear transformation from
RGB to XYZ color space first, then followed by one of two non-linear
transformations to CIE 1976 (L*u*v*) and CIE 1976 ( L*a*b*) respectively
[1].
L* = 116(Y/Y 0 )1/3 – 16 for Y/Yo > 0.008856
L* = 903.3(Y/Y0 ) for Y/Y0 <= 0.008856
u* = 13L*(u–u0)
v* = 13L*(v - v 0)
u = 4X /(X +15Y +3Z) u0 = 4X0 /(X 0 + 15Y 0 + 3Z 0)
v = 6Y /(X +15Y +3Z) v 0 = 6Y 0 / (X 0 + 15Y 0 + 3Z 0)
a* = 500[(X/X 0 )1 / 3 - (Y/Y 0)1 /3 ]
1/3
b* = 200[(Y/Y 0) - (Z/Z 0)1 /3 ]
Where (X0, Y0, Z0) are the tristimulus values of an appropriate reference
white, the perfect reflecting diffuser (100% diffuse reflectance at all visible
wavelengths), illuminated by CIE standard illumination (eg. D65). The
total color difference are given respectively by:
The CIE color spaces are perceptually uniform but they are inconve-
nient for existing non-linear transformation between them and RGB color
spaces.
