See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/270511438

The CIE 1997 Colour Appearance Model: CIECAM97s

Chapter · January 2001

CITATIONS READS
5 591

1 author:

Ming Ronnier Luo

Zhejiang University
935 PUBLICATIONS 12,165 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Colour emotion and colour harmony View project

Validity of Chromatic Adaptation View project

All content following this page was uploaded by Ming Ronnier Luo on 06 December 2018.

The user has requested enhancement of the downloaded file.

The Structure of the CIE 1997
Colour Appearance Model
(CIECAM97s)

M. R. Luo,* R. W. G. Hunt
Colour and Imaging Institute, University of Derby, Mackworth Rd., Derby DE22 3BL, England

Received 27 August 1997; accepted 10 November 1997

Abstract: The components comprising the CIE 1997 Col- tani, has also played a major role in advancing the general
our Appearance Model, CIECAM97s, are described, and philosophy of model building, and this is gratefully ac-
the steps needed to implement it in both forward and knowledged by the authors. The CIECAM97s model in-
reverse modes are listed. A worked example is also given. corporates features from various prior models including:
q 1998 John Wiley & Sons, Inc. Col Res Appl, 23, 138–146, 1998 that by Nayatani and his collaborators, K. Hashimoto, H.
Sobagaki, K. Takahama, and T. Yano; the RLAB model
Key words: colour appearance; CIECAM97s model; ad-
by R. S. Berns and M. D. Fairchild; the LLAB model by
aptation
M. R. Luo, M. C. Lo, and W. G. Kuo; and that by Hunt
and his collaborators, M. R. Pointer and M. R. Luo; refer-
INTRODUCTION ences to these models are made in the text where relevant.

At the CIE Expert Symposium 96 on Colour Standards

for Image Technology, held in Vienna in March 1996, 1 STRUCTURES OF COLOUR
APPEARANCE MODELS
it was agreed that the authors should examine existing
colour appearance models, try to combine the best fea- Colour appearance models usually comprise three stages:
tures of these models into a high-performance model for a chromatic adaptation transform, a dynamic response
general use, and test its performance against available function, and a colour space for representing the corre-
experimental data. It was also agreed that this model lates of the percepts. A simple chromatic adaptation trans-
should be available in a comprehensive version, and in a form can be provided by a multiplicative normalization
related simple version for use in limited conditions. At of X, Y, Z tristimulus values to make them constant for
its meeting held in Kyoto in May 1997, CIE Technical whites in different conditions; this was proposed by Ev-
Committee TC 1-34 agreed to adopt as the simple version ans, 2 and is incorporated in the CIELAB system.3 A more
the model CIECAM97s described in this article; the way fundamental approach is to carry out the normalization
in which this version might be extended to provide a on cone responses, as first proposed by Von Kries, 4 and
comprehensive version, CAM97c, is also described, but most recent models use modified Von Kries transforms;
this version has not yet been considered by the CIE. It is however, the Bradford transform, 5 which is used in the
important to note that these models are empirical. The LLAB model, 6 depends on a set of spectral sensitivities
inclusion of the year 97 in the designations is intended that are sharpened, compared to cone responses, and have
to indicate the interim nature of the models, pending the some small negative lobes (as shown in Fig. 1); this
availability of better models, which are expected to transform also incorporates a power function in its blue
emerge in the future. channel. The dynamic response functions used are cube-
The CIECAM97s model depends on the work of many root in the RLAB 7 and LLAB 6 models, hyperbolic in the
different investigators, and acknowledgments are made Hunt model, 8,9 and logarithmic in the Nayatani model 10 ;
to them in the text. One of these investigators, Y. Naya- these different functions diverge most for very light and
for very dark colours. The colour spaces used in these
models are similar in that they all provide approximate
*Correspondence to: Professor M. R. Luo correlates of redness–greenness and yellowness–blue-
q 1998 John Wiley & Sons, Inc. ness, ratios of which are used to derive angular correlates

138 CCC 0361-2317/98/030138-09 COLOR research and application

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
 have the same input into the transform, and, therefore,
the same output from it, apart from differences associated
with the luminance factors.
The transform converts the tristimulus values, X, Y, Z,
of test colours in a set of test conditions to tristimulus
values, Xc , Yc , Zc , of corresponding colours, which define
stimuli that have the same appearance in a set of reference
conditions. Because it is necessary to model reduced lev-
els of chromatic adaptation, the reference conditions must
include a reference chromaticity relative to which test
conditions can be considered more or less chromatic; this
reference chromaticity is chosen (as in the current RLAB
model 7 ) to be the same as that of the Equi-energy Stimu-
lus, SE , because, as has been found by Hurvich and Jame-
FIG. 1. Full lines: spectral sensitivities used in the Bradford son, 16 such stimuli appear achromatic to the dark-adapted
chromatic adaptation transform. Broken lines: spectral sen- eye. This reference chromaticity is assigned to a reference
sitivities for cones, similar to those found by Estévez. Both white. The Y value ascribed to this reference white, Ywr ,
sets of curves are linear transformations of the xV ( l ), yV ( l ),
zV ( l ) functions.
does not affect the corresponding colours produced by the
transform; this is because the reference-white parameters
used in the transform, Rwr , Gwr , Bwr , are derived from the
ratios of its tristimulus values, and these are equal to
the ratios of its chromaticity coordinates. However, it is
of hue, and square-roots of the sum of the squares of convenient to put Ywr Å 100, so that the reference white
which are used to derive correlates of chroma, while non- is the perfect diffuser. But, when using the model to obtain
linear functions of an achromatic signal are used to derive the correlate of lightness, this is evaluated relative to the
correlates of lightness; but only some of the models also corresponding colour (in the reference condition) of the
provide correlates of saturation, colourfulness, and bright- adopted test white; this is because, if evaluated relative
ness. The differences in the chromatic adaptation trans- to the perfect diffuser, the adopted white would usually
forms, in the dynamic response functions, and in the col- have a lower lightness and this would imply that it was
our spaces, all contribute to the differences between the not a true white. (In most applications in the surface
predictions made by the models. colour industries, Yw , the Y value of the adopted white,
In the CIECAM97s model, the chromatic adaptation is set to 100; this results in Ywc , the Y value of its corre-
transform is the Bradford transform (devised by Lam and sponding colour, also being equal to 100.)
Rigg 5 ), in which is incorporated a factor, D (similar to The factor, D, in the chromatic adaptation transform,
those suggested by Fairchild 7 and by Nayatani 11 ) to regu- is normally derived as a function of the luminance of the
late the extent to which the chromatic adaptation occurs adapting field, LA , which results in D increasing (and,
(as used in the LLAB model 7 ). A matrix then converts hence, the degree of chromatic adaptation increasing) as
the variables to cone responses (following the work of LA increases. But, if the colour of the illuminant is com-
Estévez 12 ), which are used as the basis of the rest of the pletely discounted (complete chromatic adaptation), D is
model; the matrix is normalized so that the cone responses set equal to 1.0; if there is partial discounting, D can be
are equal for the Equi-energy Stimulus, SE . The cone set half way between its variable value and 1.0. If there
responses are multiplied by a luminance-level adaptation is no chromatic adaptation, D is set equal to zero.
factor, 8 and then modified by a dynamic response function The nonlinear function in the B channel of the trans-
that is hyperbolic (as introduced by Seim and Valberg, 13 form means that special procedures have to be used when
and as found in physiological studies of rhesus monkeys the B signal happens to be negative.
by Valeton and Van Norren 14 ). The colour space used is
similar to that used in the Hunt model, 8 but with pre-
dictors for lightness and brightness that avoid the possibil- DYNAMIC RESPONSE FUNCTION
ity of negative values occurring (as achieved by Naya- The hyperbolic nature of the dynamic response function
tani 10 and Fairchild 6 in the Nayatani and RLAB models, results in the response gradually reducing to a noise level
respectively). Allowance for the effects of different sur- of 1.0 at very low stimulus intensities, and gradually ris-
rounds is based on the work by Bartleson and Breneman.15 ing to a maximum of 41 at very high stimulus intensities,
with a transition in between that approximates a square-
root function. The ratio of maximum to minimum re-
CHROMATIC ADAPTATION TRANSFORM
sponse, 41:1, was chosen to accord approximately with
The Bradford chromatic adaptation transform uses as in- the ratio of the maximum to minimum useful numbers of
put the ratios X/Y, Y/Y, Z/Y, instead of X, Y, Z, and this nerve impulses per second, this being estimated at about
means that a series of colours of the same chromaticity 400 to about 10.

Volume 23, Number 3, June 1998 139

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
 The luminance-level adaptation factor, FL , is propor- ing is negligible, that there is no significant contribution
tional to the cube-root of the luminance of the adapting from the rods, and that there is no low-luminance tritano-
field, LA , down to about cone threshold (5LA equal to pia. For adapting luminances in the normal photopic
about 1 cd/m 2 ) and this provides partial luminance-level range, LA equal to values between about 2 and about
adaptation over this range (full luminance-level adapta- 2000 cd/m 2 , these assumptions are likely to be valid. The
tion would correspond to FL being constant); below cone simple version also assumes that there is no significant
threshold, FL is proportional to LA , and this then provides Helson–Judd effect, and this is probably valid for most
no luminance-level adaptation, as is appropriate. typical ‘‘white’’ illuminants, but would not be true for
highly chromatic illuminants. The simple version also
makes no allowance for the Helmholtz–Kohlrausch ef-
COLOUR SPACE
fect, although this will be significant for colours of high
The colour space includes an eccentricity factor to allow chroma, except in the case of yellows.
for the fact that perceived achromatic colours are not at
the center of contours of low saturation (as in the Hunt 8,9
OUTPUT DATA
and Nayatani 10 models).
The formula for predicting chroma is constructed to The model gives predictions for hue, h and H; lightness,
allow for the fact that, although dark backgrounds make J; brightness, Q; saturation, s; chroma, C; and colourful-
most colours appear of higher chroma, in the case of light ness, M. The two predictors for hue are hue-angle, h, and
colours the opposite effect occurs so that they appear of hue quadrature, H; from hue quadrature the percentages
lower chroma (as reported by Hunt 8 ). of the unique hue components, the hue composition, can
be readily determined, and this is useful for visualizing
the perceived hue. However, hue composition has a very
INPUT DATA
nonuniform relationship with hue difference, because of
The input data to the model includes the luminance, LA , the considerably larger number of perceived hue-differ-
of the test adapting field in cd/m 2 . LA can usually be ence increments between blue and red than between red
taken as Lw /5, where Lw is the luminance in cd/m 2 of the and yellow, or between yellow and green, or between
perfect diffuser in the test illuminant; this is why 5LA is green and blue; hue-angle represents perceived hue-dif-
often used in the model instead of LA (if the adopted ferences more uniformly, but is less useful for visualizing
white is a Lambertian diffuser, Lw may be calculated as the perceived hue.
EYw / p, where E is the test illuminance in lux). Also
required are the chromaticities and luminance factors of
DETAILS OF THE MODEL
the following: the sample in the test conditions; the
adopted white in the test conditions; the background in the The steps required to use the simple version of the model
test conditions; and the reference white in the reference (CIECAM97s) are given in Appendix A, and in reverse
conditions. In addition, it is necessary to know whether mode in Appendix B; a worked example is given in Ap-
the surround conditions are average, dim, dark, or ‘‘cut- pendix C.
sheet’’; by average is meant that the surround luminance
is similar to the average luminance of the colours in the
REVERSING THE MODEL
viewing field, as is typically the case when surface colours
are viewed; by dim is meant that the surround luminance For some practical applications it is necessary to be able
is appreciably less than the average luminance of the to operate the model in reverse. The steps required to do
viewing field, as is typically the case when viewing televi- this for the simple version are given in Appendix B. It is
sion; by dark is meant that the surround luminance is to facilitate reversal that, in the formula for saturation, s,
very low compared to the average luminance of the view- the term R *a / G *a / B *a used in the Hunt94 model has
ing field, as is typically the case when viewing film pro- been replaced by R *a / G *a / (21/20)B *a .
jected in a darkened room; by ‘‘cut-sheet’’ is meant the The Bradford transform does not reverse exactly by
conditions typical for viewing cut-sheet film on light simple means because of the exponent in the B channel;
boxes. If the samples subtend more than 47 at the eye, but the procedure given in Appendix B is accurate enough
this also needs to be known. (These details, concerning for all practical purposes.17
the surround and the angular subtense, determine the val- The comprehensive version of the model, details of
ues of the parameters, F, c, FLL , and Nc , used in the which are given in Appendix D, is difficult to reverse
model.) It is also necessary to know if the colour of the because of the inclusion of the rod response; methods of
illuminant is being completely or partially discounted (in dealing with this situation are described in Appendix E.18
which case D is set at 1.0 or at an intermediate value),
or if there is no chromatic adaptation (in which case D
VISUAL AREAS IN THE OBSERVING FIELD
is set at zero).
The simple version of the model assumes that the con- For related colours, five different visual fields are recog-
ditions are such that the amount of cone pigment bleach- nized in the model:

140 COLOR research and application

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
• Colour element considered: typically a uniform patch nances divided by 2.26 are equal to photopic luminances.
of about 27 angular subtense. If LAS is not known, it can be approximated by using
• Proximal field: the immediate environment of the col-
LAS /2.26 Å LA (T/4000 0 0.4) 1 / 3 ,
our element considered, extending typically for about
27 from the edge of the colour element considered in where T is the correlated colour temperature of the illumi-
all or most directions. nant. If the scotopic luminance of the sample relative to
• Background: the environment of the colour element that of the adopted white, S/Sw , is not known, the equiva-
considered, extending typically for about 107 from the lent photopic values, Y/Yw , can be used instead.
edge of the proximal field in all, or most directions. The comprehensive version described above has not
When the proximal field is the same colour as the back- yet been considered by the CIE.
ground, the latter is regarded as extending from the
edge of the colour element considered.
MODEL FOR UNRELATED COLOURS
• Surround: the field outside the background.
• Adapting field: the total environment of the colour ele- Unrelated colours are those that are seen in isolation from
ment considered, including the proximal field, the back- other colours. Their distinguishing feature is that they are
ground, and the surround, and extending to the limit of incapable of possessing the attribute of greyness, so that
vision in all directions. they have brightnesses but no lightnesses, and colourful-
nesses and saturations, but no chromas. Light sources are
The visual patterns of scenes viewed in practice are al- frequently perceived as unrelated colours.
most infinitely variable, but the phenomenon of colour There is at present no agreed model for unrelated col-
constancy indicates that the effects of this variety on col- ours. Pending the availability of anything better, use can
our appearance are, to some extent, limited. The regime be made of the model for unrelated colours associated
of fields described above is an attempt to simplify the with the Hunt94 model.22
situation sufficiently to make it feasible for modeling,
while making it possible to include the most important APPENDIX A: STEPS FOR USING THE
factors that affect colour appearance. The proximal field CIE 1997 SIMPLE COLOUR APPEARANCE
is not used in the simple version of the model; it is in- MODEL, CIECAM97s
cluded to facilitate future incorporation in the comprehen-
sive version of the effects of simultaneous contrast, or of
assimilation for very small stimuli. Starting data:
If the colour considered has an angular subtense of
more than 47, the colorimetric measures used in the model Sample in test conditions x y Y
must be for the CIE 1964 Supplementary Colorimetric Adopted white in test
Observer, and the symbols representing the output data conditions: xw yw Yw
of the model must all carry a subscript 10. Otherwise, the Background in test
CIE 1931 Standard Colorimetric Observer is used. conditions: xb yb Yb
Reference white in
1 1
reference conditions: xwr Å 3 ywr Å 3 Ywr Å 100
Luminance of test
COMPREHENSIVE VERSION
adapting field (cd/m2) LA†
OF THE MODEL, CAM97C
† 1
In Appendix D an outline is given of the way in which LA is normally taken as 5 of the luminance of the
a comprehensive version of the model, CAM97C, could adopted test white.
be developed from the simple version, CIECAM97s. Fac-
tors are included for cone pigment bleaching, for the Hel- Surround parameters: F c FLL Nc
son–Judd effect, for the rod contribution to the achro- Average (with sample over 47) 1.0 0.69 0 1.0
matic signal, for low-luminance tritanopia, and for the Average 1.0 0.69 1.0 1.0
Helmholtz–Kohlrausch effect. Also required, but not yet Dim 0.9 0.59 1.0 1.1
implemented, are means for incorporating the effects of Dark 0.9 0.525 1.0 0.8
the proximal field on simultaneous contrast, and on assim- Cut-sheet 0.9 0.41 1.0 0.8
ilation (for stimuli of very small angular subtense).
The allowance for the Helmholtz–Kohlrausch effect is
based on the experimental results published by Fairchild
Background parameters:
and Pirrotta.19 – 21
The rod contribution is calculated in the same way Nbb Å Ncb Å 0.725(1/n) 0.2
as in the Hunt94 model 8 ; instead of using the scotopic
z Å 1 / FLLn 1 / 2 ,
luminance, LAS , of the adapting field, LAS /2.26 is used;
because, for the Equi-energy Stimulus, SE , scotopic lumi- where n Å Yb /Yw .

Volume 23, Number 3, June 1998 141

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
If the test background chromaticity is different from that 0.98699 00.14705 0.15996
of the adopted test white, then, in the above expressions, where M 01
BFD Å 0.43231 0.51836 0.04929
the Y tristimulus values of the corresponding colours in 00.00853 0.04004 0.96849
the reference conditions, Ybc and Ywc , have to be used
instead of Yb and Yw . 0.38971 0.68898 00.07868
Step 1. For the sample, calculate: and MH Å 00.22981 1.18340 0.04641 .
0.00000 0.00000 1.00000
X Å xY/y, Y, Z Å (1 0 x 0 y)Y/y and X/Y, Y/Y, Z/Y
Step 6. Calculate
and

SD S D
Ybc Å (0.43231Rbc / 0.51836Gbc / 0.04929Bbc )Yb,
R X/Y
G Å MBFD Y/Y , Ywc Å (0.43231Rwc / 0.51836Gwc / 0.04929Bwc )Yw ,
B Z/Y
n Å Ybc /Ywc, Nbb Å 0.725(1/n) 0.2, Ncb
Å 0.725(1/n) 0.2 .
0.8951 0.2664 00.1614
where MBFD Å 00.7502 1.7135 0.0367 . (If xb Å xw and yb Å yw , then Rb Å Rw and, hence, Rbc Å
0.0389 00.0685 1.0296 Rwc , and similarly for G and B, so that Ybc /Ywc Å Yb /Yw ).

Similarly, from xw , yw , Yw calculate Rw , Gw , Bw Step 7. Calculate

from xb , yb , Yb calculate Rb , Gb , Bb R *a Å 40(FL R * /100) 0.73 /[(FL R * /100) 0.73 / 2] / 1
from xwr , ywr , Ywr calculate Rwr , Gwr , Bwr .
G *a Å 40(FLG * /100) 0.73 /[(FLG * /100) 0.73 / 2] / 1
Step 2. Calculate the degree of chromatic adaptation,
D: B *a Å 40(FL B * /100) 0.73 /[(FL B * /100) 0.73 / 2] / 1.
D Å F 0 F/[1 / 2(L 1A/ 4 ) / (L 2A /300)].
Similarly, from R *w , G *w , B *w calculate R *aw , G *aw , B *aw .
But if the chromatic adaptation is complete (illuminant If R * is less than 0 use: R *a Å 040( 0FL R * /100) 0.73 /
colour completely discounted), put D Å 1.0; or if there [( 0FL R * /100) 0.73 / 2] / 1 and similarly for R *aw , and
is partial discounting put D Å 12{1 / F 0 F/[1 / for the G and B equations.
2(L 1A/ 4 ) / (L 2A /300)]}; or if there is no chromatic adapta- Step 8. Calculate:
tion, put D Å 0.
Redness–Greenness a Å R *a 0 12G *a /11 / B *a /11
Step 3. From R, G, B calculate the corresponding tristi- Yellowness–Blueness b Å (1/9)(R *a / G *a 0 2B *a )
mulus values Rc , Gc , Bc for the sample under the reference Hue angle h Å arctan(b/a).
conditions:
Rc Å [D(Rwr /Rw ) / 1 0 D]R Step 9. Using the following unique hue data:

Gc Å [D(Gwr /Gw ) / 1 0 D]G Red Yellow Green Blue

Bc Å [D(Bwr /B pw ) / 1 0 D]É BÉp
h 20.14 90.00 164.25 237.53
(when B is negative, Bc must be made negative), e 0.8 0.7 1.0 1.2

where p Å (Bw /Bwr ) 0.0834 .

Similarly, from Rw , Gw , Bw calculate Rwc , Gwc , Bwc calculate e Å e1 / (e2 0 e1 )(h 0 h1 )/(h2 0 h1 ),
from Rb , Gb , Bb calculate Rbc , Gbc , Bbc . where e1 and h1 are the values of e and h, respectively,
for the unique hues having the nearest lower value of h;
Step 4. Calculate FL Å 0.2k 4 (5LA ) / 0.1(1 0
and e2 and h2 are the values of e and h, respectively, for
k 4 ) 2 (5LA ) 1 / 3 , where k Å 1/(5LA / 1).
the unique hues having the nearest higher value of h.
Step 5. Calculate
Calculate the hue quadrature:
R* RcY H Å H1 / 100[(h 0 h1 )/e1 ]/
G* Å MH M 01
BFD GcY
B* BcY [(h 0 h1 )/e1 / (h2 0 h)/e2 ],
where H1 is 0, 100, 200, or 300 according to whether red,
R *w RwcYw yellow, green, or blue, respectively, is the hue having the
G *w Å MH M 01
BFD GwcYw , nearest lower value of h.
B *w BwcYw Calculate the Hue Composition, HC ,
where HP is the part of H after its hundreds digit, if:
142 COLOR research and application

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
H Å HP , the Hue Composition is Step 3. Using H, determine h1 , h2 , e1 , e2 ,
where e1 and h1 are the values of e and h, respectively,
HP Yellow, 100 0 HP Red, for the unique hues having the nearest lower value of h;
H Å 100 / HP , the Hue Composition is and e2 and h2 are the values of e and h, respectively, for
the unique hues having the nearest higher value of h.
HP Green, 100 0 HP Yellow, Step 4. Calculate h:
H Å 200 / HP , the Hue Composition is h Å [(H 0 H1 )(h1 /e1 0 h2 /e2 ) 0 100h1 /e1 ]/
HP Blue, 100 0 HP Green, [(H 0 H1 )(1/e1 0 1/e2 ) 0 100/e1 ],
H Å 300 / HP , the Hue Composition is where H1 is 0, 100, 200, or 300 according to whether red,
yellow, green, or blue, respectively, is the hue having the
HP Red, 100 0 HP Blue.
nearest lower value of h.
Step 10. Calculate
Step 5. Calculate e Å e1 / (e2 0 e1 )(h 0 h1 )/(h2 0 h1 ),
A Å [2R *a / G *a / (1/20)B *a 0 2.05]Nbb where e1 and h1 are the values of e and h, respectively,
for the unique hues having the nearest lower value of h;
Aw Å [2R *aw / G *aw / (1/20)B *aw 0 2.05]Nbb .
and e2 and h2 are the values of e and h, respectively, for
Step 11. Calculate the unique hues having the nearest higher value of h.

J Å 100(A/Aw ) cz , where z Å 1 / FLLn 1 / 2 . Step 6. Calculate C:

Step 12. Calculate C Å M/F L0.15 .

Q Å (1.24/c)(J/100) 0.67 (Aw / 3) 0.9 . Step 7. Calculate s:

Step 13. Calculate s Å C 1 / 0.69 /[2.44(J/100) 0.67 n (1.64 0 0.29 n )] 1 / 0.69 ,

s Å [50(a 2 / b 2 ) 1 / 2100e(10/13)Nc Ncb ]/ where n Å Ybc /Ywc .

[R *a / G *a / (21/20)B *a ], Step 8. Calculate a and b:

C Å 2.44s 0.69 (J/100) 0.67 n (1.64 0 0.29 n ), a Å s(A/Nbb / 2.05)/

M Å CF L0.15 . {[1 / (tan h) 2 ] 1 / 2[50 000eNc Ncb /13]

/ s[(11/23) / (108/23)(tan h)]}.
APPENDIX B: STEPS FOR USING THE CIECAM97s In calculating [1 / (tan h) 2 ] 1 / 2 , the result is to be taken
MODEL IN REVERSE MODE as: positive if h is equal to or greater than 0 and less than
Starting data: Q or J, M or C, H or h. Also required: 90; negative if h is equal to or greater than 90 and less
Qw , Aw , and Ybc /Ywc , obtained by using the model for- than 180; negative if h is equal to or greater than 180
ward with the adopted test white and test background. (If and less than 270; positive if h is equal to or greater than
xb Å xw and yb Å yw , then Ybc /Ywc Å Yb /Yw .) 270 and less than 360.

Surround parameters used: F, c, FLL , and Nc . b Å a(tan h).

Luminance-level parameters used: LA , D.
Step 9. Calculate
Unique hue data:
R *a Å (20/61)(A/Nbb / 2.05)
Red Yellow Green Blue
/ (41/61)(11/23)a / (288/61)(1/23)b,
h 20.14 90.00 164.25 237.53
G *a Å (20/61)(A/Nbb / 2.05)
e 0.8 0.7 1.0 1.2
0 (81/61)(11/23)a 0 (261/61)(1/23)b,

Step 1. From Q obtain J: B *a Å (20/61)(A/Nbb / 2.05)

J Å 100(Qc/1.24) 1 / 0.67 /(Aw / 3) 0.9 / 0.67 . 0 (20/61)(11/23)a 0 (20/61)(315/23)b.

Step 2. From J obtain A: Step 10. Calculate

A Å (J/100) 1 / cz Aw , where z Å 1 / FLL (Ybc /Ywc ) 1 / 2 . FL R * Å 100[(2R *a 0 2)/(41 0 R *a )] 1 / 0.73 ,

Volume 23, Number 3, June 1998 143

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
 FLG * Å 100[(2G *a 0 2)/(41 0 G *a )] 1 / 0.73 , (Y/Yc ) 1 / p B Å [É(Y/Yc )BcÉ] 1 / p /[D(Bwr /B pw ) / 1 0 D] 1 / p .
If (Y/Yc )Bc is negative, (Y/Yc ) 1 / p B is negative.
FL B * Å 100[(2B *a 0 2)/(41 0 B *a )] 1 / 0.73 .
Step 14. Calculate
If R *a 0 1 is less than 0, use Y * Å 0.43231Y R / 0.51836Y G / 0.04929(Y/Yc ) 1 / p BYc
and
FL R * Å 0100[(2 0 2R *a )/(39 / R *a )] 1 / 0.73 ,
(Y * /Yc ) ( 1 / p0 1 ) .
and similarly for the G and B equations. Step 15. Calculate

S D S D
Divide each by FL to obtain R *, G *, B *.
X 9 /Yc (Y/Yc )R
Step 11. Calculate Y 9 /Yc ÅM 01
BFD (Y/Yc )G .
Z 9 /Yc (Y/Yc ) 1 / p B/(Y * /Yc ) ( 1 / p0 1 )
RcY R*
G cY Å MBFD M 01 G* , Step 16. Multiply each by Yc to obtain X 9, Y 9, Z 9 equal
H
BcY B* to X, Y, Z, to a very close approximation.
Note: Y * differs from Y because, instead of YB, (Y/
1.91019 01.11214 0.20195 Yc ) 1 / p BYc is used: but this is multiplied by 0.04929 so
where M 01 that the difference is small. The term (Y/Yc ) 1 / p B/(Y * /
H Å 0.37095 0.62905 0.00000 .
0.00000 0.00000 1.00000 Yc ) ( 1 / p0 1 ) Å (Y/Yc )B(Y/Y * ) ( 1 / p0 1 ) ; because Y and Y * are
similar, and p is not usually very different from 1.0 (for
Step 12. Calculate Illuminant A, p Å 0.914), this term is approximately
equal to (Y/Yc )B, which is what is required to give the
Yc Å 0.43231RcY / 0.51836GcY / 0.04929BcY correct values of X/Yc , Y/Yc , and Z/Yc .
and, hence, APPENDIX C: WORKED EXAMPLE FOR THE
CIE 1997 SIMPLE COLOUR APPEARANCE
(Y/Yc )Rc , (Y/Yc )Gc , (Y/Yc )Bc . MODEL, CIECAM97s
Step 13. Calculate The CIECAM97s model gives the following results for
a sample in Standard Illuminant A (SA ) at four different
(Y/Yc )R Å (Y/Yc )Rc /[D(Rwr /Rw ) / 1 0 D],
levels of adapting luminance, LA .
(Y/Yc )G Å (Y/Yc )Gc /[D(Gwr /Gw ) / 1 0 D], Starting data:

Sample in test conditions x Å 0.3618 y Å 0.4483 Y Å 23.93

Adopted white in test conditions: xw Å 0.4476 yw Å 0.4074 Yw Å 90.0
Background in test conditions: xb Å 0.4476 yb Å 0.4074 Yb Å 18.0
Reference white in reference conditions: xwr Å 13 ywr Å 13 Ywr Å 100
Luminance of test adapting field (cd/m2), LA 2000 200 20 2
Surround: average (small sample): F Å 1.0 c Å 0.69 FLL Å 1.0 Nc Å 1.0
D factor dependent on LA

Predictions for the adopted white: Predictions for the sample:

LA 2000 200 20 2 LA 2000 200 20 2

Hue angle, h 41.8 57.4 58.8 59.5 Hue angle, h 190.2 190.0 183.5 175.7
Hue Quad., H 28.3 50.0 52.0 53.0 Hue Quad., H 239.7 239.4 229.9 218.2
Hue Comp., HC 28Y72R 50Y50R 52Y48R 53Y47R Hue Comp., HC 40B60G 39B61G 30B70G 18B82G
Lightness, J 100.0 100.0 100.0 100.0 Lightness, J 53.0 48.2 45.2 44.2
Brightness, Q 70.1 52.7 37.9 26.8 Brightness, Q 45.8 32.3 22.3 15.5
Saturation, s 0.0 0.5 12.6 25.9 Saturation, s 120.0 125.9 114.0 96.5
Chroma, C 0.1 1.3 12.1 19.8 Chroma, C 52.4 53.5 49.5 44.0
Colourfulness, M 0.1 1.3 10.8 15.7 Colourfulness, M 58.8 53.5 44.1 34.9

144 COLOR research and application

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
The saturation, chroma, and colourfulness of the white of the adopted white, S/Sw , is not known, the equivalent
increase as the adapting luminance falls, because the de- photopic value, Y/Yw , can be used instead. If the scotopic
gree of chromatic adaptation decreases; compared to the luminance, LAS , of the adapting field is not known, it can
chroma, the colourfulness increases less, because it is be approximated by using
reduced by the falling adapting luminance.
LAS /2.26 Å LA (T/4000 0 0.4) 1 / 3 ,
where T is the correlated colour temperature of the illumi-
APPENDIX D: STEPS FOR USING THE 1997
nant.
COMPREHENSIVE COLOUR APPEARANCE
MODEL, CAM97c Calculate
The steps required for the comprehensive version of the A Å [2R *a / G *a / (1/20)B *a / AS 0 2.31]Nbb .
model, CAM97c, are the same as those for the simple
version, CIECAM97s, given in Appendix A, except as Calculate Aw similarly.
indicated below.
Step 13. Ft is a low-luminance tritanopia factor:
Step 7. Calculate
Ft Å LA /(LA / 0.1)
R *a Å BR {40(FL R * /100) 0.73 / bt Å bFt
[(FL R * /100) 0.73
/ 2] / R *D } / 1, s Å [50(a 2 / b 2t ) 1 / 2100e(10/13)Nc Ncb ]/
G *a Å BG {40(FLG * /100) 0.73 / [R *a / G *a / (21/20)B *a ].
[(FLG * /100) 0.73
/ 2] / G *D } / 1, Step 14. Helmholtz–Kohlrausch effect added:

B *a Å BB {40(FL B * /100) 0.73 / JHK Å J / (É100 0 JÉ)(C/300)(sinÉ((h 0 90)/2)É),

[(FL B * /100) 0.73 / 2] / B *D } / 1, QHK Å (1.24/c)(JHK /100) 0.67 (Aw / 3) 0.9 .

where BR , BG , BB are cone bleach factors:

APPENDIX E: METHODS OF REVERSING
BR Å 10 7 /[10 7 / 5LA (R *w /100)] THE COMPREHENSIVE VERSION
OF THE MODEL, CAM97c
BG Å 10 7 /[10 7 / 5LA (G *w /100)]
Reversing the comprehensive version of the model,
BB Å 10 7 /[10 7 / 5LA (B *w /100)]. CAM97c, is complicated by the presence of the rod con-
tribution in the achromatic signal.
If no cone bleaching occurs, these factors are all equal
The value of AS depends on S/Sw ; this is not usually
to 1.
known, and Y/Yw is usually used in the model instead as
R *D , G *D , B *D are Helson–Judd effect factors: an approximation. But, when using the model in reverse,
R *D Å kD (R * /R *w 0 Ybc /Ywc )(R * /R *w 0 1/3), if Y/Yw is not known, some extra steps are necessary.
There are two alternative methods of dealing with this
G *D Å kD (G * /G *w 0 Ybc /Ywc )(G * /G *w 0 1/3), situation. The first method is to use J to calculate (S/Sw )J ,
an approximate value for S/Sw . The following formula for
B *D Å kD (B * /B *w 0 Ybc /Ywc )(B * /B *w 0 1/3), (S/Sw )J can be used for photopic conditions for colours
where kD is a constant, the value of which is adjusted to of luminance factor not less than 3%:
provide the appropriate amount of Helson–Judd effect. (S/Sw )J Å {J 0 [log(5LA )] 2 /4} 1.8 /
If there is no Helson–Judd effect, or if the colour of the
illuminant is discounted, R *D Å G *D Å B *D Å 0. {100 0 [log(5LA )] 2 /4} 1.8 .

Step 10. Calculate the rod contribution, AS : (S/Sw )J can then be used in the reverse model to calculate
X, Y, and Z. Using (S/Sw )J instead of Y/Yw in the forward
AS Å BS (3.05)(FLS S/Sw ) 0.73 /[(FLS S/Sw ) 0.73 / 2] / 0.3, model results in H and s being unchanged, C and M
where BS Å 0.5/{1 / 0.3[(5LAS /2.26)(S/Sw )] 0.3 } being changed very slightly, and Q and J being changed
slightly. For many applications these changes are negligi-
/ 0.5/{1 / 5[5LAS /2.26]} ble. If more precise results are required, the resulting
2
and FLS Å 3800 j 5LAS /2.26 value of Y/Yw can be used as S/Sw to derive a new set of
values of X, Y, and Z; this procedure can then be iterated
/ 0.2(1 0 j 2 ) 4 (5LAS /2.26) 1 / 6 until stable values of Y/Yw are obtained. The second
method is to require that S/Sw be equal to Y/Yw from
and j Å 0.00001/(5LAS /2.26 / 0.00001).
the outset, and to use methods of successive numerical
If the scotopic luminance of the sample relative to that approximation to complete the calculation.

Volume 23, Number 3, June 1998 145

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B 1940
 1. CIE Publ. No. x010-1996, Proceedings of the CIE Expert Sympo- 11. Y. Nayatani, Examination of adaptation co-efficients for incomplete
sium ’96 Colour Standards for Image Technology, Central Bureau chromatic adaptation. Color Res. Appl. 22, 156–164 (1997).
of the CIE, Vienna, 1996. 12. O. Estévez, On the fundamental data-base of normal and dichro-
2. R. M. Evans, Visual processes and color photography. J. Opt. Soc. matic colour vision, Ph.D. Thesis, University of Amsterdam, 1979.
Am. 33, 579–614 (1943). 13. T. Seim and A. Valberg, Towards a uniform color space: a better
3. CIE Publ. No. 15.2, Colorimetry, 2nd Ed., Central Bureau of the formula to describe the Munsell and OSA color scales. Color Res.
CIE, Vienna, 1986. Appl. 11, 11–24 (1986).
4. J. von Kries, Die Gesichtsempfindungen, in Handbuch der Physi- 14. J. M. Valeton and D. van Norren, Light adaptation of primate cones:
ologie des Menschen, Vol. III, Braunschweig, Vieweg, 1905, pp. an analysis based on extracellular data. Vision Res. 23, 1539–1547
109–282. (1983).
5. K. M. Lam, Metamerism and colour constancy, Ph.D. Thesis, Uni- 15. C. J. Bartleson and E. J. Breneman, Brightness perception in com-
versity of Bradford, 1985. plex fields. J. Opt. Soc. Am. 57, 953–957 (1967).
16. D. Jameson and L. M. Hurvich, Use of spectral hue-invariant loci
6. M. R. Luo, M. C. Lo, and W. G. Kuo, The LLAB(l; c) colour
for the specification of white stimuli. J. Exp. Psychol. 41, 455–463
model. Color Res. Appl. 21, 412–429 (1996).
(1951).
7. M. D. Fairchild, Refinement of the RLAB color space. Color Res.
17. R. W. G. Hunt, Reversing the Bradford chromatic adaptation trans-
Appl. 21, 338–346 (1996). form. Color Res. Appl. 22, 355–356 (1997).
8. R. W. G. Hunt, An improved predictor of colourfulness in a model 18. R. W. G. Hunt, The Reproduction of Colour, 5th Ed., Fountain
of colour vision. Color Res. Appl. 19, 23–26 (1994), and The Press, Kingston-upon-Thames, England, 1995, pp. 753–754.
Reproduction of Colour, 5th Ed., Fountain Press, Kingston-upon- 19. M. D. Fairchild and E. Pirrotta, Predicting the lightness of chromatic
Thames, England, 1995. colors using CIELAB. Color Res. Appl. 16, 385–393 (1991).
9. R. W. G. Hunt and M. R. Luo, Evaluation of a model of colour 20. Y. Nayatani, Comments on predicting the lightness of chromatic
vision by magnitude scalings: discussion of collected results. Color object colors. Color Res. Appl. 17, 412–416 (1992).
Res. Appl. 19, 27–33 (1994). 21. M. D. Fairchild, Fairchild replies. Color Res. Appl. 17, 416–417
10. Y. Nayatani, H. Sobagaki, K. Hashimoto, and T. Yano, Field trials (1991).
of a nonlinear color-appearance model. Color Res. Appl. 22, 240– 22. R. W. G. Hunt, Measuring Colour, 2nd Ed., Ellis Horwood, Chich-
258 (1997). ester, 1991, pp. 251–257.

146 COLOR research and application

/ 8A1B$$1940 04-30-98 12:27:05 crra W-Color Res

8A1B
View publication stats 1940