A Color Correction System using a Color Compensation Chart

Seok-Han Lee, Tae-Yong Kim, and Jong-Soo Choi

Dept. of Image Engineering, Graduate School of Advanced Imaging Science, Multimedia, and
Film Chung-Ang University, DongJak-Ku, Seoul 156-756 Republic of Korea
{ichthus, tykim, jschoi}@imagelab.cau.ac.kr
Abstract
In this paper, we propose a color correction system
using a color compensation chart. The implementation
target of the proposed system is the color image
captured by digital imaging device, and the system
introduces the color compensation chart designed for
it to estimate the transformation between the colors in
the image and the reference colors. The color
correction process of the proposed system consists of
two steps, i.e., the profile creation process and the
profile application process. During the profile creation
process, the relationships between the captured colors
and the reference colors are estimated. And the system
creates a color profile and embeds the estimation
result. During the profile application process, the
colors in the images which are captured under the
same condition as that of the chart image are
corrected using the created color profile. To evaluate
the performance of the system, we perform experiments
under various conditions. And we compare the results
with those of widely used commercial applications.

1. Introduction
With the advance of digital imaging devices such as
digital cameras, their abilities to represent accurate
colors have been important issues. However, accurate
color representation is a difficult problem for digital
imaging devices, because color images of unknown
objects are captured under various unknown conditions
such as illuminations. Furthermore, color distributions
of captured images are also dominated by
characteristics of the imaging devices.
In this paper, we propose the color correction
system using the color compensation chart. The
implementation target of the proposed system is the
color image captured by digital imaging device such as
digital camera or digital scanner. The proposed system
uses the color compensation chart designed for it to

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

Fig. 1. The block diagram for the proposed system.

estimate the relationship between the colors in the
image and the reference colors. The color correction
process consists of two steps, i.e., the profile creation
process and the profile application process. During the
profile creation process, the relationships between the
captured colors and the reference colors are estimated.
And the system creates a color profile and embeds the
estimation result in the profile. During the profile
application process, the colors in the images which are
captured under the same condition as that of the chart
image are corrected using the created color profile.

2. System description
We adopted the concepts of color management
system (CMS) and profile connection space (PCS) to
realize the proposed system. Fig.1 shows the block
diagram of the proposed system [2]. First, the image of
the color compensation chart is captured. And a
relationship between the colors of the patches in the
captured chart image and the internal reference colors
is estimated. After the estimation, the system creates an
ICC color profile, and embeds the relationship in the
profile. The created profile is used as the source profile
as in Fig.1 to correct the colors of the other images

(a) The color compensation chart. (b) Color gamuts considered to

determine the patch colors.
Fig. 2. The color compensation chart and its gamut.

Fig. 3. Indices of the patches.

which are captured under the same condition as that of

the chart image.

3. Color compensation chart for the

proposed system
For the proposed system, we designed the color
compensation chart as shown in Fig.2(a). This chart
provides the reference colors for the estimation of the
relationship between the colors in the image and the
actual colors. By the relationship, every color in the
captured image is transformed into the actual color.
Eight patches located in the center area of the chart are
used to create TRCs. The leftmost three patches in the
compensation chart provide the primary colors of RGB.
The colors of the patches were determined based on
the most general colors observed in common
environment and general skin colors of Korean people
Table.1. Patch colors.
L*

71.18

89.39

70.58

43.35

31.98

71.18

a*

-0.37

6.63

22.04

32.32

19.46

-0.37

b*

-0.37

6.61

11.04

19.83

6.13

-0.37

L*

38.00

87.14

84.24

54.57

41.52

42.49

a*

43.71

1.74

10.57

5.95

5.90

11.50

b*

20.96

27.87

52.94

48.67

24.54

18.55

L*

56.08

88.11

71.28

55.22

36.40

44.94

-12.54

-24.66

-24.62

-13.43

-17.49
-0.29

a* -37.07
b*

15.32

2.66

16.32

8.28

5.23

L*

39.58

95.50

88.70

81.15

62.53

47.85

a*

1.60

0.23

-0.04

-0.78

-0.13

40.08
36.62

b* -35.32
L*

60.05

a* -16.50

0.25

-0.01

-0.28

-0.39

79.51

40.96

33.70

23.12

53.35

0.66

-0.61

0.38

0.14

-16.33

b*

25.63

0.43

-0.40

2.96

-0.25

-5.03

L*

52.33

86.78

77.20

63.91

34.27

83.54

a*

41.92

-4.71

-10.11

-13.44

-10.63

-10.32

b*

-1.53

-0.60

11.32

-7.68

-16.13

-9.65

L*

53.54

88.13

80.36

50.60

29.96

38.07

a*

5.95

1.86

4.88

25.15

12.62

-6.50

b*

77.19

0.59

-7.76

-20.52

-17.88

-15.08

L*

71.18

80.54

74.88

67.73

43.10

71.18

a*

-0.37

9.34

10.64

12.13

24.62

-0.37

b*

-0.37

15.13

18.04

19.17

26.85

-0.37

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

Fig. 4. Profile creation process.

[6, 7]. The color gamuts of widely used digital cameras
were also considered, because the proposed system is
for the images captured by digital cameras. Fig.2(b)
shows the color gamuts of the compensation chart and
the considered digital cameras. The + marks in
Fig.2(b) denote the patch colors, and they were
adopted to be located in the considered gamuts. Table
1 shows the L*a*b* values of the patches in the chart,
and Fig.3 shows the index of every patch which is used
in the entire process.

4. Color correction algorithm for the

proposed system
4.1. Profile creation process.
Fig.4 shows the block diagram of the color profile
creation process. The first step of the profile creation

1.2

1
0.9

Normalized Color //

0.7
0.6
0.5
C aptu r ed
Fi tte d
Refe rnc e

0.4
0.3
0.2

Outp ut C o lo r (No rm aliz ed )///

0.8

0.8

0.6
Red
Green
Blue

0.4

0.2

0.1

0
(4, 4)

(4, 3)

(4, 2)

(4, 1)

(3, 4)

(3, 3)

(3, 2)

(3, 1)

Index of Patch

(a) Result of curve regression.

0.2

0.4
0.6
0.8
Input Color (Normaliz ed)

1.2

(b) TRCs for each channel.

Fig. 5. Results of the TRC estimation process.

process is the TRC estimation. And the color in the
RGB space is converted into that of CIEXYZ color
space. Then the transformation to map the color to the
correct color is estimated. The results of the
estimations are embedded in the ICC profile.
4.1.1. Estimation of the tone reproduction curves.
The eight patches located in the center area of the chart
are use for the TRC estimation process (i.e., from (3,
1) to (3, 4) and from (4, 1) to (4, 4) in Fig.3). For each
of R, G, and B channels, a function to fit the color
values of these eight patches in the image to those in
the internal reference chart is estimated, and the
function is used as the TRC. This process is performed
in R, G, and B channels independently using the colors
of the eight patches in each channel, and we get three
TRCs during this process. In this paper, we apply
curve regression to fit the color values of the eight
patches to those of the internal reference chart as
follows [11]:
g ( x ) = a1 f1 ( x ) + a 2 f 2 ( x ) + a3 f 3 ( x ) + a4 f 4 ( x ), 0 x 1,
f 1 ( x ) = 1, f 2 ( x ) = x, f 3 ( x ) = sin( x ), f 4 ( x ) = exp( x ),

(1)

where an is undetermined coefficient, and x is

normalized color of the patch to be in the range of 0.0
to 1.0. The coefficients in Eq.(1) are determined to
minimize the sum of the square of the differences
between the fitted color values of the eight patches in
the image and those of the internal reference chart.
2

R = [ ri ] = yi a n f n ( xi ) ,
i =1
i =1
n =1

(2)

This estimation is performed for each of R, G, and

B channels, and we get three TRCs. Fig.5 shows the
result of the TRC estimation process. Fig.5(a) is the

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

result of the curve regression, and Fig.5(b) shows three

TRCs for each of R, G, and B channels. Regression
errors of the patches used for the regression are shown
in Table 1. The TRCs are preserved in the created ICC
profile for the profile application process as described
in section 4.1.4.
4.1.2. Color space conversion. As described in
section 2, the system adopts PCS and CMS. They were
defined by ICC to provide a reference color space as a
common interface for color specification between
different devices. Accordingly, every process in the
PCS is independent of kinds of devices [2]. And the
color space such as CIEXYZ or CIELAB is employed
in the device independent color space. In the proposed
system, XYZ color space was introduced for the PCS.
Therefore, RGB colors of the patches in the image
should be converted into those of XYZ color space
before the estimation of the transformation. However,
there exist some conditions which should be
considered for the color space conversion. As defined
in the specification ICC.1:2004-10 [2], the PCS is
relative to a specific illuminant (i.e., D50). Moreover,
we should consider the connection between the PCS
and the device color space. Therefore, following
conditions are considered to estimate the color
conversion matrix:
1) Reference white of the PCS: The PCS defined
in the specification ICC.1:2004-10 is based on
the illuminant D50. Therefore, the reference
white point in the PCS should be equal to D50
(i.e., X=0.9642, Y=1.0, Z=0.8249).
2) Device color space: We employed sRGB as the
device color space. The sRGB space is the
standard device color space for most of digital

imaging or displaying devices. It is defined

based on the illuminant D65.
3) Chromatic adaptation transformation: As the
illuminant of the device color space is not equal
to that of the PCS, the chromatic adaptation
method is required for the color space
conversion. We used the linear Bradford model
for the chromatic adaptation transformation,
which is recommended by the specification
ICC.1:2004-10 for the chromatic adaptation
transformation [1, 2].
By the condition 1), 2), we get the color space
conversion matrix which maps the color in the RGB
space to that of the XYZ space follows:
0.412424 0.357579 0.180464
M cvt = 0.212656 0.715158 0.072186 .
0.019332 0.119193 0.950444

Table 3. ICC profile Tags for the system.

(3)

And by the condition 3), we can get the chromatic

adaptation transformation matrix Madapt as follors:
M adapt

1.047835 0.022897 0.050147

= 0.029556 0.990481 0.017056 .
0.009238 0.015050 0.752034

(4)

By Eq.(3), and Eq.(4), the color space conversion

matrix MCVS for the proposed system can be estimated
as follows:
0.436052
M CVS = M adapt M cvt = 0.222492
0.013929

0.385082
0.716886
0.097097

0.143087
0.060621 .
0.714185

(5)

4.1.3. Correction of the image color in the PCS. The

colors of the image are corrected by the linear
transformation which maps the colors of the image into
those of the reference colors in the PCS. The
transformation is estimated as follows:
M XYZ = M Crr M' XYZ
XN
X1 X 2
= Y1 Y2 L Y N
Z 1 Z 2
Z N
m11 m12 m13 X '1
= m 21 m 22 m 23 Y '1
m 31 m 32 m 33 Z '1

(6)
X '2
Y '2
Z '2

Fig. 6. Profile Structure.

X 'N
Y ' N ,
Z ' N

where Xi, Yi, and Zi denote the colors of the reference

chart, and Xi, Yi, and Zi are those of the color patches
in the chart image. MCrr is the transformation matrix

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

Tag

Signature

Data Type

mediaWhitePointTag

wtpt

XYZType

mediaBlackPointTag

bkpt

XYZType

CopyrightTag

cprt

multiLocalizedUnicodeTag

profileDescriptionTag

desc

multiLocalizedUnicodeTag

redMatrixCouilmnTag

rXYZ

XYZType

greenMatrixCouilmnTag

gXYZ

XYZType

blueMatrixCouilmnTag

bXYZ

XYZType

redTRCTag

rTRC

curveType

greenTRCTag

gTRC

curveType

blueTRCTag

bTRC

curveType

which maps the colors of the image into those of the

reference colors. Eq.(6) is rearranged to be estimated
estimated by SVD (singular value decomposition) as
follows:
X 1 X '1
X X'
2= 2
M

X N X 'N
Y1
Y
2
M

Y N

X '1
X'
= 2


X 'N

Z1
Z
2
M

Z N

X '1
X'
= 2


X 'N

Y '1
Y '2
M
Y 'N
Y '1
Y '2
M
Y 'N

Y '1
Y '2
M
Y 'N

Z '1
m11
Z ' 2
m ,
12

m13
Z 'N

(7)

Z '1
m 21
Z ' 2
m ,
22
m 23
Z 'N

(8)

Z '1
m 31
Z ' 2
m .
32

m 33
Z 'N

(9)

4.1.4. ICC color profile created by the system. The

estimated relationship to correct the image color is

embedded in the ICC color profile. The ICC color

profile should include the necessary elements as single
file. Fig.6 shows the structure of the ICC profile.
Table.3 shows the tags used for the ICC profile created
by the system. They are defined in the specification
ICC.1:2004-10 [2]. The leftmost column represents the
tag type, and the middle column demotes the signature
embedded in the profile to specify the tag type. The
rightmost column means the data type for each tag type
defined in the specification.
1) mediaWhitePointTag, and mediaBlackPointTag:
The reference white and the reference black are
encoded in each tag. The chromatic adaptation
transformation is applied to the color space
conversion matrix in the proposed system. And the
ICC profile prescribes that the mediaWhitePointTag
values should be equal to the illuminant D50 for
display
devices.
Therefore,
the
mediaWhitePointTag of the proposed system is set
to the values of D50. The mediaBlackPointTag is
set to the reference color of black (i.e., X = 0, Y = 0,
Z = 0).
2) redMatrixColumnTag, greenMatrixColumnTag,
and blueMatrixColumnTag: These tags are
intended to be used for the transform from the
device color space to the PCS. In the proposed
system, the color space conversion matrix
multiplied by the color correction matrix is
embedded in each of redMatrixColumnTag,
greenMatrixColumnTag,
and
blueMatrixColumnTag as follows:
t11
M T = M Crr M CVS = t 21
t 31

t12
t 22
t 32

t13
t 23 ,
t 33

redMatrixC olumn X t11

redMatrixC olumn = t ,
Y

21
redMatrixC olumn Z t31
greenMatri xColumn X t12
greenMatri xColumn = t ,
Y

22
greenMatri xColumn Z t32

input color normalized to be in the range of 0.0 to

1.0 to the output color in the RGB color space as
follows:
redTRC [i ] = g R ( x i ),
g reen TRC [i ] = g G ( x i ),
b lue TRC [i ] = g B ( x i ), i = 0, 1, 2, ..., N S , x i =

i
.
NS

(12)

Here, NS denotes the number of samples of g(x) which

is estimated by Eq.(1). In the proposed system, g(x) is
sampled at 1024 points in the range of 0.0 to 1.0. The
subscripts R, G, and B represent the g(x) values in each
of R, G, and B channel, respectively.

4.2 Profile application process

In the profile application process the image colors are
corrected by the relationship embedded in the color
profile. Fig.7 shows the block diagram of the profile
application process [2]. The tone reproduction is
applied to the color of the input image. And the color
is corrected by the matrix embedded on the created
color profile.

(10)
Fig. 7. Profile Application process.

5 Experimental results
(11)

blueMatrix Column X t13

blueMatrix Column = t .
Y

23
blueMatrix Column Z t33

3) rTRCTag, gTRCTag, and bTRCTag: The TRCs

for each of R, G, and B channels are preserved in
each of rTRCTag, gTRCTag, and bTRCTag. The
proposed system employs the curveType data type
defined in the ICC specification. For the curveType,
a 1-D lookup table(LUT) is established to map the

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

The experiments are performed in the two

categories. First, the performance of the system is
evaluated under the various light sources. We use
SpectraLight III by GretagMacbeth which can produce
CIE standard light sources. The second experiment is
performed for the white balance distortions. For the
second experiment, we distort the white balance of the
captured image with the preset modes of the digital
camera used for the experiment.
The performance of the system is evaluated by the
average of the color difference of every patch. The
color difference of every patch is estimated as follows:

Table 4. The color differences for the various

illumination models.
D65

CWF

Horizon

U30

"A"

Original

4.24

8.1

28.4

17.6

21.9

Proposed

5.82

6.91

14.7

9.01

QP card

4.32

5.38

17.6

7.32

Table 5. Comparison of the color differences for the

white balance modes.
Sunlight

Shadow

Cloud

Original

6.94

16.82

9.39

31.2

22.41

8.51

Proposed

5.87

6.57

7.59

9.62

9.21

7.92

QP card

8.44

8.72

8.57

11.71

11.52

5.95

6.44

7.68

15.53

14.21

6.13

6.89

7.45

16.29

15.82

Profile
Maker 5

3.44

4.42

9.57

6.23

6.88

Profile
Maker 5

i1

3.86

4.77

12.4

8.75

9.05

i1

Fig. 8. Comparison of the performances of the

systems for the illumination models.

E*ab = (( L*) + ( a*) + ( b*) )

2

2 1/ 2

,
L* = L *1 L *2 , a* = a *1 a *2 , b* = b *1 b *2 , ,

(13)

where L*1, a*1, and b*1 represent the captured color,

and L*2, a*2, and b*2 mean the internal reference color.
The color difference of every patch is averaged, and
we consider this average of the color difference as the
color correction error. The performance of the system
for each experiment is compared with those of three
commercial applications: ProfileMaker 5 and i1 by
GretagMacbeth, QP Color Kit 1 by QPcard. These
systems have their own color charts and their own
unique profile formats. The images are captured by
Canon EOS-10D digital camera.
5.1. Experimental results for the various light
sources

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

Tungsten Fluorescent

Fig. 9. Images for the comparison of the color

differences for the white balance distortions.
For the experiments, we consider the CIE standard
illumination models generated by SpectraLight III:
D65 (6500K), CWF(4150K), Horizon(2300K),
U30(3100K), and Standard light A(2856K). The
images are captured under these illumination models,
and the color differences are estimated in CIELAB
space for the evaluation of the system performance.
However, every system considered for the experiments
has its own color chart. Therefore, we employ
ColorChecker color chart by GretagMacbeth as the
test chart, which is not used by any systems. The
experiment is performed as follows. First, the image of
each systems native color chart is captured under the
illumination model. And each systems color profile
for the illumination model is created for the color
correction. Then, the image of the test chart of Fig.10
is captured under the same illumination model as that
of the native color chart image of each system. The
image of the test chart is corrected with the color
profile which was created based on the image of the
native chart. Finally, the color correction error of each

Captured

Corrected

Captured

(a)

Corrected

(b)

Fig. 10. The experimental results under the arbitrary white balance modes.
Captured

Corrected

(a)

Captured

Corrected

(b)

Fig. 11. The experimental results under the arbitrary conditions.

system for the test chart image is estimated, and
compared.
The estimated color differences are shown in Table
4. From the result, it seems that ProfileMaker5 shows
the best results. However, we can verify that the
proposed system presents almost equivalent
performance to the other commercial systems. Fig.8
shows the image of the experimental results.
5.2 Experimental results for the white balance
distortions
The white balance of the captured image is distorted
with the preset modes of the digital camera. The digital

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

camera used for the experiment can be set to the preset

white balance values such as 1) Sunlight (5200K:
X=0.9605, Y=1.0, Z=0.8636), 2) Shadow (7000K:
X=0.9494, Y=1.0, Z=1.1597), 3) Cloud (6000K:
X=0.9523, Y=1.0, Z=1.0080), 4) Tungsten (3200K:
X=1.0, Y=1.0, Z=0.2605), and 5) Florescent (4000K:
X=0.9963, Y=1.0, Z=0.6095). The experiment is
performed in the similar manner as that of section 5.1.
However, the white balance value is distorted instead
of the illumination variation.
Table 5 represents the color differences for the
white balance distortions. From Table 5, we verify that
the proposed system is not much affected by the white
balance distortions. Moreover, it shows the superior

performance to the other systems. Fig.9 shows the

results for the white balance distortions.
5.3. Experimental results for the visibility
Fig.10 and Fig.11 show the experimental results for
the arbitrary conditions. The purpose of this
experiment is to ensure the visibility presented to the
users. Fig.10 presents the results for the arbitrary white
balance modes. The results show that the color
correction performance of the proposed system is not
much affected by the white balance distortions.
Fig.11(a) shows the examples for the outdoor images.
The colors of the captured images were distorted by
Fluorescent mode under unknown illuminations. We
can verify that every color becomes clearer than that of
the original image. The color correction performance is
also verified by the white color and the green tone
reflected on the window in the image. Fig.11(b)
presents another examples, which also confirms the
color correction performance of the proposed system.

6. Conclusion
We proposed the color correction system using the
color compensation chart designed for it. To correct
the color of the captured image, we employed the
concepts of CMS and PCS. And we introduced the
ICC color profile to embed the estimation results for
the color correction in it. The color corrections for the
images which were captured under the same condition
as that of the chart image were performed with the ICC
color profile created by the proposed system. The
proposed system shows almost equivalent performance
to the commercial systems under the various
illumination models. For some white balance
distortions, however, we verified that the proposed
system provides superior performances. The
experimental results present the possibility of the
proposed system for the practical use. Through the
experiments, we found that the errors of the results
may be reduced more by refinement processes. In our
future work, the refinement process to reduce the error
will be considered. And a process which can
compensate the error caused by non-uniform
brightness of the chart image will be included. The
experiments were performed with digital cameras.
However, the proposed system may be used for other
digital imaging devices such as digital photo-scanner.
The modification for other digital imaging devices also
will be considered in our future work.

7. Acknowledgments

2006 International Conference on Hybrid Information Technology (ICHIT'06)

0-7695-2674-8/06 $20.00 2006

This work is supported by the Ministry of

Education and Human Resource Development under
BK21 project, and financially supported by the
Ministry of Education and Human Resource
Development (MOE), the Ministry of Commerce,
Industry and Energy (MOCIE) and the Ministry of
Labor (MOLAB) through the fostering project of the
Lab and Excellency.

8. Reference
[1] Hsien-Che Lee, Introduction to Color Imaging
Science, Cambridge University Press, 2005.
[2] Specification ICC.1:2004-10 (Profile version
4.2.0.0), International Color Consortium, 2004.
[3] Hung-Shin Chen and Hiroaki Kotera, Threedimensional Gamut Mapping Method Based on
the Concept of Image Dependence, Journal of
Image Science and Technology, Vol.46, No.1,
pp44-52, Jan./Feb. 2002.
[4] B. Pham and G. Pringle, Color Correction for an
Image Sequence, IEEE Computer Graphics and
Applications, pp38-42, 1995.
[5] Eung-Joo Lee, Favorite Color Correction for
Reference Color, IEEE Transactions on
Consumer Electronics, Vol.44, No.1, Feb., 1998.
[6] Byung-Tae Ahn, Eun-Bae Moon, and Kyung-Suk
Song, Study of Skin Colors of Korean Women,
Proceedings of SPIE, Vol.4421, pp705-708, June
2002.
[7] Jung-Sook Jun, A Study on the Color Scheme of
Urban Landscape Based on Digital Image Color
Analysis, Yonsei Graduate School of Human
Environmental Sciences, 2002.
[8] H.C. Do, S. I. Chien, K. D. Cho, H. S. Tae, Color
Reproduction Error Correction for Color
Temperature Conversion in PDP-TV, IEEE
Transactions on Consumer Electronics, Vol. 49,
No. 3, pp. 473-478, Aug. 2003.
[9] D. I. Han, Real-Time Color Gamut Mapping
Method for Digital TV Display Quality
Enhancement, IEEE Transactions on Consumer
Electronics, Vol. 50, No. 2, pp. 691-699, May
2004.
[10] B. Pham and G. Pringle, Color Correction for an
Image Sequence, IEEE Computer Graphics and
Applications, pp38-42, 1995.
[11] S. Nakamura, Applied Numerical Methods in C,
Prentice Hall, London, 1995.