September 10, 2009 / Vol. 7, No. 9 / CHINESE OPTICS LETTERS 873
An adaptively spatial color gamut mapping algorithm
Xiandou Zhang (
ÜÜÜ
www
ÌÌÌ
) and Haisong Xu (
MMM
°°°
ttt
)
∗
State Key Laboratory of Modern Optical Instrumentation, Zhejiang University, Hangzhou 310027, China
∗
E-mail: chsxu@zju.edu.cn
Received December 10, 2008
To improve the accuracy of color image reproduction from displays to printers, an adaptively spatial color
gamut mapping algorithm (ASCGMA) is proposed. In th is algorithm, the compression degree of out-
of-reproduction-gamut color is not only related to th e position of the color in CIELCH color space, but
also depending on the neighborhood of the color to be mapped. The psychophysical experiment of pair
comparison is carried out to evaluate and compare this new algorithm with th e HPMINDE and SGCK
gamut mapping algorithms recommended by the International Commission on Illumination (CIE). The
experimental results indicate that the proposed algorithm outperforms the algorithms of HPMINDE and
SGCK except for the very dark images.
OCIS codes: 330.1715, 330.1730, 330.5510, 110.3000.
doi: 10.3788/COL20090709.0873.
When a color image is reproduced from one digital device
to another, the color appearance of the reproduction im-
age will disagree with the origina l one if there are colors
in the image being out of the reproduction device gamut,
which usually occurs in the image reproduction from
computer displays to printers
[1]
. Herewith, gamut map-
ping is an important issue in ima ge reproduction, and has
been one of the most active directions of color manage-
ment research. A number of gamut mapping algorithms
have been developed in these years, for which Moroviˇc has
made a survey
[2]
. As a whole, the color gamut mapping
algorithms can b e classified into three categories. The
first ca tegory is the “device-to-device” gamut mapping
algorithm, which is a function of the input and output
gamuts
[3]
. The majority of well-known gamut mapping
algorithms fall in this category. The second category is
the “image-to-device” gamut mapping algorithm, which
is the function of the image statistics
[4−6]
. Both the
first and second categories belong to the “point-to-point”
type ga mut mapping algorithms, which do not consider
the neighbor r e lationships in the image s. Accordingly,
the spatial gamut mapping algorithms, classified into the
third categor y, were brought o n rece ntly
[7−12]
. For such
gamut mapping algo rithms, the mapped color is not only
dependent upon the output gamut, but als o upon the
colors of the neighbor points in the or iginal image.
There is not yet a sta ndard gamut mapping algo-
rithm though a plethora of algorithms have been devel-
oped. Up to now, the International Commission on Illu-
mination (CIE) has only recommended two a lgorithms,
the hue-angle preserving minimum ∆E
∗
ab
clipping (HP-
MINDE) and chroma-dependent sigmoid lightness map-
ping and cusp knee scaling (SGCK) algorithms
[13]
. The
HPMINDE algorithm is the simplest gamut mapping al-
gorithm, which maps the out-of-reproduction-gamut col-
ors to the boundary of the reproduction gamut with mini-
mum color difference, while the colors in the reproduction
gamut are preserved. This algorithm can maximally pre-
serve the chroma of the original colors, but it can cause
unacceptable artifacts as most of the out-of-gamut colors
may be mapped to the same point. The SGCK algorithm
is a combination of GCUSP (chroma-dependent lightness
compression and linear compression to cusp)
[2]
and SLM-
CKS (sigmoid lightness mapping and cusp knee scaling)
algorithms
[14]
. The lightness range of the original device
gamut is firstly scaled to the lightness range of the repro-
duction gamut through the chroma-dependent sigmoid
function, then the lightness and chroma of the lightness
scaled color is mapped to the reproduction gamut simul-
taneously while preserving the hue, where the lightness
of anchor point is set as the lightness of the cusp on the
reproduction gamut boundary, and the knee line is set
as 90% of the reproduction gamut boundary in reference
to the anchor point. For the SGCK a lgorithm, there are
three main shortcomings though it performs well
[15]
and
is recommend by CIE. Firstly, the lightness of anchor
point is set as the lightness of the cusp on the reproduc-
tion gamut boundary, which indeed can make the best
use of the reproduction gamut. But if the lightness of
the cusp on the reproduction gamut boundary is very
low or high, there will be too much compression for the
out-of-gamut colors. Secondly, the knee line is heuris-
tically set as 90% of the reproduction gamut boundary.
The main adva ntage of this setting is that the process
sp e ed is very fast, but if most of the image colors are
out of the 90% of reproduction gamut, these colors will
be compres sed to the remaining 10% of the reproduc-
tion ga mut, which results in the loss of image details,
and so the artifacts will appear even worse. Thirdly, the
SGCK algorithm does not take the neighborhood pix-
els color into account, which also causes spatial infor-
mation loss. Most of the existing spatial gamut map-
ping algor ithms decompose the original image into sev-
eral spatial frequency bands, then apply different gamut
mapping strategies for the individual spatial frequency
bands, finally the proce ssed bands are summed together
with different weights to derive the mapped color
[10,11]
.
This kind of algorithms could indeed keep the spatial re-
lationship of the original image in the reproduction one.
However, as the dealing strategies are different among the
individual spatial frequency bands, an effective merging
of the processed bands is not an easy work
[11]
, which will
affect the mapping qua lity of the algorithms. In a ddi-
tion, the mapping speed tends to be slow as different
1671-7694/2009/090873-05
c
2009 Chinese Optics Letters