Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

48th International Symposium ELMAR-2006, 07-09 June 2006, Zadar, Croatia

Comparison of Color Difference Equations

Lidija Mandic I, Sonja Grgic 2, Mislav Grgic 2


'University of Zagreb, Faculty of Graphic Arts, Getaldiceva 2, 10000 Zagreb, Croatia
2
University of Zagreb, Faculty of EE & Comp, Unska 3/XII, 10000 Zagreb, Croatia
E-mail: lidUa.mandic@,grf hr

Abstract: Advanced color difference equations still do not necessarily predict accurately the color differences
for complex stimuli such as photographic images. The CIELAB color-difference formula has been used
extensively for quality control in industry but its application is limited because CIELAB space is stillfar from
being perceptually perfectly uniform. The consequence of this is that for equal perceptual color diferences, the
values of CIELAB color differences can vary by an order of magnitude. The aim of this work is to compare
accuracy of three color diference equations that are recently used.

Keywords: CIELAB Space, Color Diferences

1. INTRODUCTION b* 200[(Y/ )1/3 (z/z )113]


_ (3)
Color difference research has been active over
three decades. The aim is to develop a single number Clb = -,/at +b2 (4)
shade pass/fail equitation for evaluating the small to
medium color difference [1]. Although the system of
color specification introduced in 1931 by the CIE ab a (5)
and augmented in 1964 has served the color industry
well, there remain a number of problems. One of the In these equations X, Y and Z are the tristimulus
main problem is that in terms of visual perception values of the stimulus and X1, Yn, Zn are the
CIE system is non uniform. Equal changes in x, y do tristimulus values of the reference white. L*
not correspond to perceived differences of equal represents lightness, a* approximates redness-
magnitudes. Most attempts to develop more uniform greenness, b* approximates yellowness-blueness,
spaces have sought to find linear or non-linear C*ab approximates chroma and h*ab approximates
transforms of the tristimulus values or chromaticity hue. The CIELAB color circle is shown in Fig. 1.
coordinates to give a more uniform color space.
Between 1940 and 1976 a great number of color YelHow
spaces, were proposed as uniform color spaces [2].
b

Some of those, such as HunterLab and ANLAB,


were quite successful but in 1976 the CIE agreed
upon two transformations that led to CIELAB and
CIELUV. CIELUV was intended to be used to
specify the colors of lights and other self-luminous
sources, whereas CIELAB was intended to be used
for the specification of surface colors.
+d

2. CIELAB COLOR SPACE

Currently the most important color space based


on the opponent-color theory is known as CIELAB.
The L*, a* and b* coordinates can be calculated
from the tristimulus values XYZ normalized to the
white by equations (1)-(5): -b
Blue

P = 116 Y -16 (1) Fig. 1. CIELAB color circle


a= 500X/X )1/3 _ (y/y )1/3] (2) The chroma is varied as a parameter from the center
to the edge of the circle. The popular
representational form LCH (lightness-chroma-hue)

107
48th International Symposium ELMAR-2006, 07-09 June 2006, Zadar, Croatia

is produced from the CIELAB color circle by means acceptable) color difference. Consider drawing a
of a simple variation. In it chroma C* and hue h* are boundary around the sample that represents a JND,
determined from the values a* and b* in accordance the AE*ab draws a perfect sphere around the sample.
with Fig. 1. The goal of the CIELAB color space A more realistic model for human vision describes
design was to have perceptually uniform color color differences inside elliptical regions that differ
differences throughout the space. in the shape and size throughout the color space.
For a perfect agreement between experimental
data and CIELAB space, all ellipses should be
3. COLOR DIFFERENCE EQUATION constant size circles. However, experiments showed
different results.
Color differences are measured in the CIELAB Luo and Rigg accumulated most of the available
space as the Euclidean distance between the experimental data relating to small to medium color
coordinates for the two stimuli. This is expressed in differences of surface colors [3]. It was found that
terms of a CIELAB AE*ab that can be calculated the ellipses formed a consistent pattern in terms of
using equation (6): their shapes and orientations, but not sizes, as shown
in Fig. 3. Ellipses close to neutral colors are the
smallest. Ellipses are larger and longer when chroma
AE ab = V(AL*)2 + (Aa *)2 + (Ab *)2 (6)
is increased. Most ellipses point towards the neutral
point except for those in the blue region.
It can also be expressed in terms of lightness,
chroma and hue differences as shown in equation (7)
and Fig. 2:
.,

20~~~
AE * = V(AL*)2+ (AC *b )2+ (AH*b )2 (7) a
so 0
e
where
20 8QOZO *L0
4
000
AL* = LT -LS 2
A\a* =a* - a* 40 5 l%V
Ab* = bTT - b*s
T S 0
-2
9

AH*ab =(Aa
~ aF (Ab )2_w (AC*
*)2 + 1) abJ)2
-4 30 ~ ~
-6
60 -40 -20 0 20 40 60 80 100 120
The subscripts refer to the standard color (S) and the
trial color (T), AH*ab is the difference in hue and
AC*ab is simply the difference between C* of the Fig. 3. Luo and Rigg experimental color
standard and the trial colors. discrimination ellipses plotted in a*b* diagram
Since 1976 attempts to generate better metrics
for the prediction of color differences have
concentrated on finding more sophisticated measures
AL* of distance [4]. In 1978, CIE published guidelines to
coordinate researches studying color difference.
Berns et al. in collaboration with the DuPont
company conducted visual assessments [5]. The data
from assessments were used to derive a relatively
simple color-difference equation, named CIE94,
which was recommended for field trials by CIE in
1994. The ellipses show the same trends as found in
Fig. 2. Euclidean distance between points in Fig. 3.
CIELAB space All the advanced equations have a common
feature: they were derived by modifying the
How do we interpret the AE scale? A AE of 1 is CIELAB equation. A generic formula given in
defined as a just-noticeable difference (JND). This is equation (8) represents all these equations:
the threshold at which a trained observer would just
notice the difference between two colors. There is
the concept of perceivable (can I see a difference in AE j( AL * + (AC * AH * (8)
color) and acceptable (is this difference in color

108
48th International Symposium ELMAR-2006, 07-09 June 2006, Zadar, Croatia

where AR=RTf(AC*AH*), and AL*, AC* and AH* L' = L*


are the CIELAB metric lightness, chroma and hue
a'= (1+G) a*
differences, respectively, calculated between the
standard and sample in a pair. AR is an interactive b' =b*
term between chroma and hue differences. The SL,
Sc, and SH are the weighting functions for the c1 = a'2 + b
lightness, chroma and hue components, respectively.
The value calculated for these functions vary h = tan- '(b'/ a')
according to the position of the sample pair being The G and T terms are computed using:
considered in CIELAB color space. The kL, kc, and
kH values are the parametric factors to be adjusted
according to different viewing parameters, such as G=0,5 1-0 a2 J
textures, backgrounds, etc., for the lightness, chroma
and hue components. For CIE94, an interactive term ab +2
between the hue and chroma differences AR is set to
zero and SL equals one. The geometric mean of the T = 1- 0,17cos(h' -30)+ 0,24cos(2h')+ 0,32cos(3h'+ 6) - 0,20cos(4h'- 63°)
two samples should be used for calculating Sc and RT = -sin(2AO)Rc
SH. The CIE94 has hue weighting function SH which
is independent of hue angle. RC= 2(C *7 /(C *7 +257
The CIE94 equation is (9): AO = 30exp [h' -2750) 25TJ4

AE94= jk + (bk jAHb


+ (9) The AEOO include not only the lightness, chroma and
weighting functions, but also an interactive term
between the chroma and hue differences for
where improving the performance for blue colors and a
SC 1+0,045C*ab scaling factor for the CIELAB a* scale for
SH=1+0,015C*ab improving the performance for colors closed to the
achromatic axis [6]. It was agreed that the arithmetic
Equation (9) gives large errors in predicting mean between the standard and sample of a pair
chromatic differences for saturated blue colors and should be used for calculating all weighting
wrongly predicts chromatic differences between functions, rather than using the geometric mean or
neutral colors. the standard of a pair.
The CIE have recently recommended for trial the
CIEDE2000 (AEOO) color difference formula that is
given by equation (10): 4. EXPERIMENT
AE00 To examine variations among color difference
(10) equations, visual experiment using two most
(ks, (k SC +( k kS (A J kH H J commonly used media, CRT-displays (soft copy)
and printed images (hard copy), were performed. Six
where printed images containing pictorial information were
used as the originals. Experiment was carried out in
S =1+ 01 5(L'- 50)2 illuminated room. Printed hard copies were
20 + (L'- 50)
illuminated and viewed under light source that
simulated CIE Standard Illuminants D50. Displayed
SC = 1+ 0,045C' image had the same physical size as original image.
A total of five observers, took part in the visual
SH =1+0,015C'T experiment [7]. Observers adjusted the rendition of
the image on CRT display to match the color
The terms AL', AC' and AH' are given by appearance of the printed image. The printed
originals and CRT reproduction were used so as to
AL'=L'T-L's correspond to simultaneous binocular viewing. The
AC' =C'T-C's images were segmented into several number of
object region, depend of image context, that were
AH' = 2CCs sin(Ahi2) measured using a Gretag Spectrolino spectro-
photometer.
Ah' =h1 -h/

109
48th International Symposium ELMAR-2006, 07-09 June 2006, Zadar, Croatia

5. RESULTS The results show that AEOO gives the smallest color
differences for cyan, white and black colors. The
Color differences between printed images and AE*ab gives the largest color differences for all
images displayed on CRT for blue and neutral colors colors. The AE94 and AEOO color differences are
are shown in Table 1. similar for some blue colors, but for other blue
colors AEOO gives better results. The AEOO metric is
Table 1. Color differences changed dependent on the position in color space,
and an interactive term between chroma and hue
Color AE*ab AE94 AEoo differences that is included in the equation (10)
7,1 5,7 5,3 improves results that are achieved using this
7,2 4,9 5,5 equation.
blue 7,4 6,3 5,5
9,3 7,2 6,3
7,1 4,8 5,8 5. CONCLUSION
18,1 11,4 11
15,6 12,9 11,6 There are a number of proposed improvements to
neutral 12,2 8,9 8,2 the AE calculation. The new versions of AE are still
15,1 9,9 9,9 based on CIELAB; the only thing that has changed
21,3 14,5 15,4 is the way in which the calculation is done. The
15,9 12,7 11,4 newer measures for AE are better because they
15,3 13,3 11,8 calculate an elliptical rather than a spherical distance
neutral 13,2 10,7 10,8 in CIELAB space. The shape and size of these
17,3 13,3 12,5 ellipses change dynamically throughout the color
12 10,1 9,6 space. The appropriate method depends on
11,3 12,7 9,8 application. For matching purpose between two
10,5 13,3 8,5 different media, we recommend the use of AEOO
blue 10,3 10,7 8,6
22,4 13,3 7,9
because it provides a better agreement between
10,5 10,1 6,8 results of visual assessments and measured color
14,6 10,8 10,4 differences.
8,9 8,3 8,7
neutral 6,5 5,1 5,1 REFERENCES
9,2 8,2 7,8
12,3 10,7 10,5 [1] Colorimetry, 2nd Edition, Publication CIE
11,7 7,1 8,5 15.2-1986
13,2 8,5 9,3 [2] K. McLaren, B. Rigg, "The SDC recommended
blue 9,9 6,1 6,6
9,2 6,8 6,9
colour-difference formula: change to
CIELAB", Journal of the Society of Dyers and
8,5 5,7 5,9 Colourists, 1976, pp. 337-338
9,4 6,4 6,2
11,2 7,8 6,9 [3] M.R. Luo, B. Rigg, "Chromaticity-
cyan 13,2 6,8 4,7 discrimination ellipses for surface colours",
15,8 9,4 7,1 Color Research andApplication, 1986, pp.25-42
22,8 10,5 6,3 [4] A.R. Robertson, "CIE guidelines for
18,2 10,3 13,6 coordinated research on color difference
16,9 9,4 12,5 evaluation", Color Research and Application,
blue 15,1 9,6 10,5 1978, pp. 149-151
20,8 12,1 15,7 [5] D.H. Alman, R.S. Berns, G.D. Snyder, W.A.
16,9 9,9 11,2 Larsen, "Performance testing of color-
21,4 14,1 14,2 difference metrics using a color tolerance
17,2 10,8 10,8 dataset", Color Research and Application,
blue 23,4 12,8 10,6 1989, pp. 139-151
23,9 15,7 16,1 [6] M.R. Luo, G. Cui, B. Rigg, "The Development
28,3 14,7 11,5 of the CIE 2000 Colour-Difference Formula:
20,4 16,6 13,7
17,5 14,5 12,1 CIEDE2000, Color Research and Application,
white 18,6 15,4 12,8 Vol. 26, Issue 5, October 2001, pp. 340-350
13,2 11,3 9,8 [7] L. Mandic, S. Grgic, T. Kos "Image matching
17,9 14,8 12,4 between hard and soft copies under different
9,5 9 8,1 viewing conditions", Proceedings of the 47th
9,7 9,7 9,6 International Symposium ELMAR-2005
black 9,9 8,5 4 focused on Multimedia Systems and
14,7 14,2 9,2 Applications, Zadar, Croatia, 08-10 June 2005,
9,9 9,1 6 pp. 33-36

110

You might also like