October 10, 2007 / Vol. 5, No. 10 / CHINESE OPTICS LETTERS 617
Prediction of Chinese color system appearance scales using
various color appearance models
Yusheng Lian (
), Xiuze Wang (
), and Wei Meng (
)
Institute of Science, Information Engineering University, Zhengzhou 450001
Received January 4, 2007
The chromaticities of the Chinese color system dataset are applied to eight color appearance models
(CAMs). Models used are: CIELAB, Hunt, Nayatani, RLAB, LLAB, ZLAB, CIECAM97s, CIECAM02.
Three color appearance attributes (lightness, chroma, and hue) are discussed for their uniformity, in terms
of the constant perceptual nature of the Chinese color system dataset. The results show that no particular
model can excel at all metrics. Comparison can lead to the conclusion that Chinese color system appearance
scales can be predicted only slightly poorer than Munsell appearance scales using the eight CAMs.
OCIS codes: 330.1690, 330.1720, 330.1730.
Color appearance model (CAM) is an important tool
which solves the problem of the color fidelity display or
communication under complex illumination and viewing
conditions and between cross-media
[1]
. Hence, there is
a strong need by color imaging engineers to integrate a
CAM with color management systems. Several CAMs
have been developed and refined in recent years, each de-
rived with a different approach and stressing the various
aspects of perception to a greater or less degree. So, these
models, most widely known and used, should be tested
using some available data groups. And Chinese c olor
system data can be used to test these models because
of its uniformity. This study compares several modern
CAMs with respect to their abilities to predict uniformly
appearance scales of Chinese color system data, here-
after, compares the results of Chinese color system with
Munsell color system. Mo dels used were: CIELAB,
Hunt, Nayatani, RLAB, LLAB, ZLAB, CIECAM97s,
CIECAM02
[2]
.
Chinese color system has a structure similar to Munsell
system. It is built along the three perceptual quantities:
hue, chroma (describing saturation), and value (describ-
ing lightness). In the album, adjacent color samples
represent equal intervals of visual perception. It consists
of 1338 samples arranged on 40 pages of constant hue
[3,4]
.
The number of samples in Chinese color system is about
a half of that in Munsell system
[5]
.
Input data of these models are the chromaticity coor-
dinates of the Chinese color system data and the model-
specific parameters for viewing conditions. The concrete
parameters used to calculate model coordinates for Chi-
nese color system data sets are: CIELAB: X
n
=95.02,
Y
n
= 100, Z
n
= 108.81; Hunt: X
w
=95.02, Y
w
= 100,
Z
w
= 108.81, L
A
=63.6619, T = 6504, N
c
=1,N
b
= 75,
Y
b
= 20; Nayatani: X
n
=95.02, Y
n
= 100, Z
n
= 108.81,
Y
o
= 20, E
o
= 5000, E
or
= 1000; RLAB: X
n
=95.02,
Y
n
= 100, Z
n
= 108.81, σ =1/2.3, D = 1; LLAB:
X
n
=95.02, Y
n
= 100, Z
n
= 108.81, F
S
=3,F
L
=1,
F
C
=1,Y
b
= 20, D =1;ZLAB:X
n
=95.02, Y
n
= 100,
Z
n
= 108.81, exps = 0.345, L =63.6619, F =1;
CIECAM97s: X
w
=95.02, Y
w
= 100, Z
w
= 108.81,
L
A
=63.6619, C =0.69, N
c
=1,F =1,Y
b
= 20,
F
LL
=1,D = 1; CIECAM02: X
w
=95.02, Y
w
= 100,
Z
w
= 108.81, L
A
=63.6619, C =0.69, N
c
=1,F =1,
Y
b
= 20. Parameters were chosen to consistently and
appropriately represent the viewing conditions recom-
mended for Chinese samples: daylight (Illuminate D
65
)
and average surrounding
[6]
.
All these models predicted the perceptual color at-
tributes of lightness, chroma, and hue. These three color
appearance attributes are divided into three dimensions.
The discussion focuses on each of these separately. These
three dimensions in any model cannot be appropriately
combined. This is because that, in the original scaling
experiments of Chinese samples, observers adjust each
dimension of color separately.
The performance of mo dels’ lightness linearity is illus-
trated in Fig. 1, in which the model lightness is plotted
against Chinese value
[7]
. A good lightness sca le should
be linear with C hinese value. A linear fit is done on
each set of model lightness data. For comparison, corre-
lation coefficients for regression lines are shown in Ta ble
1. From it one can observe that the level of correlation
is high as shown by the correlation coefficient R
2
of
0.9944 or higher. Moreover, the differences of correla-
tion coefficient a mong these models are very small. It
can be confirmed by linearity of each tie-line in Fig. 1 .
The larger the value of correlation coefficient, the better
Fig. 1. Lightness linearity of the eight CAMs.
1671-7694/2007/100617-04
c
2007 Chinese Optics Letters