Methods for Measuring Viewing Parameters in CIECAM02

Chenyang Fu and M. Ronnier Luo, Department of Colour and Polymer Chemistry, University of Leeds, United Kingdom

Abstract SR =
LSW
(1)
This is the continuation of a previous study1 which investigated the LDW
possibility of applying a characterised digital camera to measure
colours and quantify viewing parameters used in colour where LSW is the luminance of surround white and LDW is the
appearance models such as CIECAM02. The earlier results luminance of media white (which can be the device white point).
showed that a characterised digital camera can measure CRT Luminance units are cd/m2.
colours with reasonable accuracy in terms of tristimulus values
under average, dim and dark surround conditions. The present There are three categories of surround ratios5, i.e. SR 0.2, SR<0.2
study investigates the effects of viewing parameters on image and SR= 0, corresponding to three types of surround, average, dim
appearance under various surrounding conditions. Six methods and dark, respectively. After the value of SR is obtained, the
were developed to determine viewing parameters such as the viewing conditions are defined and hence viewing parameters such
luminance level, the luminance factor of background and the as F (incomplete adaptation factor), c (lightness surround induction
surround conditions (two methods were described in an earlier factor) and Nc (chromatic surround induction factor) can be
paper1). The standard method was based on that recommended by determined. Values for these parameters in CIECAM02 are
CIECAM025 and measurements were conducted using a Minolta summarised in Table 1.
CS1000 tele-spectroradiometer (TSR). The performance of the
other five methods were evaluated through comparison with the Table 1: CIECAM02 Viewing Parameters and Surround Ratio SR
standard data set. Twenty four colours included in a single image c Nc F SR
were used as test colours. The results show that using an image’s Average
mean luminance value as the luminance factor of the background 0.69 1.0 1.0 0.2
surround
(Yb) led to the best agreement with the standard method. Dim
0.59 0.9 0.9 < 0.2
surround
Introduction Dark
It is well known that background and surround have certain impact 0.525 0.8 0.8 0
surround
on the appearance of an image. Many previous studies have been
published on this issue. This study aims to investigate the viewing Note that the definition of surround in CIECAM02 is not the same
parameters used in colour appearance models, which is a crucial as that in ISO 3664:20007. CIECAM02 adopts the terms
part of the colour management technology used for faithfully “surround” and “background” that were used in ISO 3664 in 1991
reproducing colour images across different media. but which are now opposite to ISO 3664:2000. The definition of
surround in ISO 3664:2000 is “the area adjacent to the border of
The structure of a colour appearance model includes three parts: a an image which, upon viewing the image, may affect the local state
chromatic adaptation transform used to predict corresponding of adaptation of the eye.” For example, in the case of a reflection
colours from one set of illumination conditions to another, dynamic copy, the border is usually taken to mean the unprinted region
response functions and a colour space.2,3 A colour appearance immediately adjacent to the image and is called the surround of the
model has the ability to describe the appearance of colours and to printed image in ISO 3664:2000, but it is called background in
ensure a colour appearance match under a given set of conditions. CIECAM02. In this study, the terms “surround” and “background”
CIECAM024, 5 was recently recommended by the CIE to industries have both been adopted using the definitions in CIECAM02. It
as an international standard colour appearance model, with the should be noted that this confused situation might cause difficulties
following definitions: in applying colour appearance models.

• Adapting field  everything in the visual field outside of the Our previous study1 showed that determining the colour and
stimulus. luminance levels of background and surround could be a difficult
• Background  a roughly 10 degree region immediately task for complex images presented under realistic viewing
surrounding to the stimulus. environments when using a conventional TSR. Therefore a
• Surround  the field outside the background6. characterised digital camera was used with both the earlier and
present studies to collect colour data for a given viewing
A property of surround, the surround ratio (SR), is found thus: environment.

13th Color Imaging Conference Final Program and Proceedings 69

It should be noted that the luminance factor of background (Yb) A viewing environment was created with a reference white located
depends on the image contents. Green8 found that although this in the adapting field. This was regarded in this viewing environment
effect is not significant, there is still scope for further investigation. as the surround white. The CRT peak white was treated as the
device white. The entire viewing environment is illustrated in
The aim of this study is to investigate the effects of various viewing Figure 1.
parameters on the colour appearance of colour patches and pictorial
images on a CRT under different surround conditions. To
investigate how much the viewing parameters contribute to the
perceived match under different viewing conditions, existing colour
appearance model were evaluated to find ways of improving them.

Viewing parameters in CAMs include illuminant, luminance level,

luminance of adaptation field (LA), luminance factor of background
(Yb) and surround. This study focuses on methods for estimating LA
and Yb. Six methods were developed to obtain these viewing
Figure 1. The experimental set up
parameters and CIECAM02 was used to compare the effect.
Viewing parameters were obtained using a Minolta CS1000 TSR, a
camera characterisation model, a weighting method for viewing
Methods for Quantifying Viewing Parameters
field (Yb) and a low-pass image method.
Each of the colour patches in the digital 24-colour chart was
presented in the centre of the CRT monitor (as shown in Fig 1) and
Experimental Conditions measured by using the TSR. The device white (CRT peak white)
The experiment was conducted in a laboratory in which a CIE LDW, and surround white (the reference white in adapting field) LSW
Illuminant D65 simulator with dimmer control was used to set up were measured using the TSR. The surround ratio SR was
the desired viewing environment. In order to minimise flare
calculated using Eq (1). Measured LDW, LSW and calculated SR in the
reflected off the CRT display, a black cloth covered objects in front
real experiment are summarised in Table 2. For each surround
of the monitor (behind the TSR/camera). The flare on the CRT
condition, three sets of tristimulus values for each colour patches Xi
display was measured was and found to be around 0.15 cd/m2
Yi Zi ( i = 1 to 24 ) were measured using the TSR and the luminance
under average viewing conditions. This value was less than 0.5% of values for the adapting field (LA) and luminance factor of
the CRT maximum luminance. The viewing distance was 70 cm
background (Yb) were estimated in six different ways.
from the CRT display according to the office working environment.
A camera-characterisation model was developed with which images
Table 2: LSW, LDW and Surround Ratio SR Used in the
were taken to determine the viewing parameters under different
Experiment under Three Surround Conditions
surround conditions.
LDW LSW SR
Average 68.07 23.3 0.342
Experimental Set-up
Dim 67.85 4.34 0.064
A 24-bit graphic card, an HP P1100 CRT monitor and a 6.1
Dark 66.65 0.10 0.001
effective mega pixel Nikon D1X digital camera were used. The
CRT monitor was adjusted to a CCT of 6500K. The luminance of
the CRT’s white point was set to around 67 cd/m2 which was Note that it has been suggested4 that “for a self-luminous display,
obtained from the results of testing monitor’s channel and spatial the device white point refers to the colour generated by setting each
independence9. of the self-luminous primaries to the maximum possible value.”
Therefore, the device white used in this study was set with a square
The CRT monitor and digital camera were carefully characterised peak white patch (digital counts R=G=B=255) displayed in the
as described in the earlier studies.1 Note that the camera centre of the HP P1100 monitor with a black colour (digital counts
characterisation model developed using polynomial regression10 R=G=B=0) as background.
was derived using digital colour patches displayed on the CRT.
These colour patches were all uniform colours generated from the Method 1: In CIECAM025, it is recommended that the
captured colour chart in a viewing cabinet illuminated by a D65 chromaticity and luminance value of the monitor peak white should
simulator. Adobe Photoshop was used to obtain the average RGB be measured using a TSR. LA (the luminance value of adapting
values for each colour patch. Using these average RGB values, field) was calculated by Eq. (2).
digital uniform colour patches were created. A digital colour chart
with 240 colour patches was used as training data for camera Yb1 Lbackground
characterisation1. A digital 24-colour chart with 24 colour patches LA1 = LDW × when Yb1 = × 100 (2)
100 LDW
generated from a captured GretagMacbeth ColorChecker 24 was
used as the test colours in this study.

70 Society for Imaging Science and Technology & Society for Information Display
where LDW is the luminance of device white (the peak white of
CRT display in this study). Luminance units are cd/m2 and Yb is the
luminance factor of background.

This method was derived from the recommendation of CIECAM025

and measurement was conducted using the TSR. The advantage of
this instrument is that the measurement results correspond to the Figure 2. The experimental set up displaying the digital 24-colour chart
actual viewing conditions. Therefore, this data set (LA1 and Yb1) was
used as the standard for comparison with the results obtained from
the other 5 methods. Method 4: The same procedure as for Method 3 was used to
capture the image of 24-colour chart which was treated as one
Method 2: The mean value of the absolute Y tristimulus value complex stimulus image. Yb4 was derived from the mean value of
(luminance channel) for the 24 digital colour patches was calculated the luminance channel of the captured image (the 24-colour chart
by: itself) using image processing algorithms at each pixel location. LA4
was then determined using Eq. (4).

Yb 2 = ∑Y 24
i
(3) Method 5: The method of using a weighting function to determine
Yb, which was developed by Green7, was used in this phase.
Weights for the captured image and for the background luminance
where Yi, is the absolute Y tristimulus value for each colour patch. values were calculated by “integrating the weighting function with
LA2 was then determined using Eq (4). the relative angular substance occupied by the fields.”8 A Gaussian
function was employed as the weighting function, i.e. greater
Ybi × L DW weight was given to the middle of the image and gradually less
L Ai = (4)
100 towards its edges. Table 3 shows the weights for the image and for
the background under a CRT viewing environment (shown in
where LAi is the luminance of adaptation field of each method i, LDW Figure 2). Yb5 was calculated by “multiplying each element in the
is the device white. Luminance values are in units of cd/m2. column of weights by the corresponding total luminance for the
field and summing these individual contributions to the total
Method 3: This method uses a camera characterisation model to background luminance.”8 LA5 was determined by Eq. (4).
predict tristimulus values of test colour patches, to predict device
white and surround white, and finally to determine the viewing
parameters for CIECAM02. This method was described in the Table 3: Weightings for Image and Background in Calculating
earlier study.1 Background Factor Yb5 Under Dark Surround Condition
Dark Flare Mean Y Distance Weight Yb5
An illustration of this viewing environment is shown in Figure 1 Background 0 1.76 0.48- 0.33 19.66
which displays a peak white patch which was first captured using 1.0
the digital camera under three surround conditions, set up by Image 0.15 27.98 0.0- 0.67
adjusting the ambient illumination. A camera-characterisation 0.48
model was developed using a polynomial regression method10.
Prediction worked well under three surround conditions. The model
was used to obtain the predicted device white (XDWp YDWp ZDWp) and Methods 6(1) and 6(2): Recently, imaging technologists and
the predicted surround white (XSWp YSWp ZSWp), from which LSWp, scientists realised the importance of the spatial components in an
LDWp were derived. Next, the surround ratio SRp was determined image11, 12, 13. It is known that sensitivity of the visual system is
using Eq. (1). different at different spatial scales. Therefore, it is possible that
using image processing algorithms to reduce the undetectable
Accordingly, a digital 24-colour chart (see Figure 2) was captured details in an image might not influence observers’ perception of the
and the camera characterisation model used to predict the image. This step might not necessarily result in lower image quality.
tristimulus values for each of the 24 colour patches Xip Yip Zip ( i = 1 Many researchers10, 11, 12 developed the contrast-sensitivity function
to 24). LA3 was determined using Eq. (4) with LDWp, and then Yb3 in the colour-difference evaluation of complex images and lately it
was obtained from Eq. (2). has been incorporated into colour appearance models. For example,
in iCAM11, 12 which was developed by M.D. Fairchild and G.M.
Johnson, the luminance of the image (low-pass filtered) and the
surround was used to modulate the exponents.

In this study, low-pass filtered images were used to investigate the

influence of different parts of an adapting field on viewing

13th Color Imaging Conference Final Program and Proceedings 71

parameters and the degree of contribution of different parts to the performed slightly better than J, ac, bc”14. Therefore, J, aM, bM were
perceived image match under various viewing conditions. The used in this study. The colour difference of each colour patch
luminance values for the adapting field (LA61, LA62) were obtained between the data measured using TSR in Method 1 and those
from low-pass images of the luminance channel. The two low-pass obtained from other methods based on camera model (excluding
filtered images used here included the captured 24-colour chart Method 2) was calculated using Eq. (5).
itself (Figure 3b) and the chart plus the surround condition (Figure
3a) respectively. The values of Yb61 and Yb62 were determined using
∆E Jab = ∆J 2 + ∆a M 2 + ∆bM 2 (5)
Eq. (2) with LA61, LA62 and LDW as the variables.

where aM = Mcos(h), bM = Msin(h) and J, M, h represent

CIECAM02 attributes lightness, colourfulness and hue angle.

Results and Discussion

To evaluate these methods, a digital colour chart was presented on
the same CRT display. The image of this viewing environment (as
shown in Figure 1) was captured using the same digital camera
under three viewing conditions (i.e. average, dim and dark).
Figure 3. Different parts of captured images in Method 6 for calculating Ymean
The viewing parameters were obtained using various methods,
including predicted data using a camera-characterisation model,
Evaluation of Different Methods calculated weights for image and background and the data from a
The six methods were evaluated by comparing the measured data low-pass version of the image. The TSR was used to measure the
(Method 1) with data calculated from each method. Note that in XYZ values for each colour patch and the luminance values for
these methods, the surround ratio was first obtained using Eq. (1) device white (LDW) and surround white (LSW), excluding Method 3
with measurement results for LDW and LSW. Next, this ratio was used in which the predicted data were used for all viewing parameters. It
to determine the viewing parameters c, Nc and F according to Table should be noted that in Method 6, the viewing parameter LA was
1. To simplify the evaluation, except for Method 31, the TSR was derived from a low-pass image of the luminance channel. The low-
used to measure the XYZ tristimulus values for each colour patch as pass filtered images were chosen from different parts of a captured
well as the luminance values for the device white (LDW) and image (shown in Figure 3 as LA61, a digital colour chart itself and
surround white (LSW). Only the viewing parameters were LA62, the chart with viewing environment together, respectively).
determined using each of the methods. Note that the fundamental The data achieved from Methods 2 to 5 were compared with those
image information was first derived from the camera derived from measurement data (LW) from the TSR (Method 1) in
characterisation model in Methods 4, 5 and 6. CIECAM02. Table 4 shows the viewing parameters, LA, Yb, LSW,
LDW and surround ratio SR, used in CIECAM02 obtained from the
The tristimulus values, XYZ, were transformed to CIECAM02 J, six methods under average surround conditions.
aM, bM colour space (based on J, M and h polar space) with
corresponding viewing parameters. In order to take into account Table 5 and 6 show the comparison results between Method 1 and
viewing conditions, the J, aM, bM colour space was used here rather the other methods in terms of colour difference in CIECAM02 J,
than CIELAB. Colour spaces such as JCh and QMh which are aM, bM colour space under average and dark surround conditions.
approximately uniform colour spaces, are used in CAMs. They are The viewing parameters used here to calculate CIECAM02 J, aM,
similar in that they all provide redness-greenness and yellowness- bM were the obtained from the five methods (Methods 2 to 5). Note
blueness scales to form rectangular coordinates (like CIELAB). that for the Method 6, when the LA was obtained from low-pass
Note that in the study14 to evaluate colour difference using colour images of a different part of the captured image (Figure 3), the
appearance models, the CIECAM02 “J, aM, bM colour space results were quite different than when under dark and under
average surround conditions.

Table 4: Viewing Parameters used in CIECAM02 Under Average Surround Conditions

Average Name LDW LSW SR LA Yb
Method1 TSR 68.07 23.3 34.23 13.61 19.7
Method2 TSR 68.07 23.3 34.23 13.61 21.67
Method3 Camera 69.19 24.6 35.33 13.84 18.6
Method4 Picture 68.07 23.3 34.23 14.03 20.61
Method5 Weights 68.07 23.3 34.23 19.04 27.97
Method6(1) Low-pass(1) 68.07 23.3 34.23 20.61 30.28
Method6(2) Low-pass(2) 68.07 23.3 34.23 16.59 24.37

72 Society for Imaging Science and Technology & Society for Information Display
Table 5: Mean and Maximum ∆EJab Values in CIECAM02 J aM bM Colour Space Between the Data Obtained from Method 1 and Those
Obtained from Other Methods Under Average Surround Conditions
Average Name LDW LSW SR LA Yb
Method1 TSR 68.07 23.3 34.23 13.61 19.7
Method2 TSR 68.07 23.3 34.23 13.61 21.67
Method3 Camera 69.19 24.6 35.33 13.84 18.6
Method4 Picture 68.07 23.3 34.23 14.03 20.61
Method5 Weights 68.07 23.3 34.23 19.04 27.97
Method6(1) Low-pass(1) 68.07 23.3 34.23 20.61 30.28
Method6(2) Low-pass(2) 68.07 23.3 34.23 16.59 24.37

Table 6: Mean and Maximum ∆EJab Values in CIECAM02 J aM bM Colour Space Between the Data Obtained from Method 1
and Those Obtained Using Other Methods Under Dark Surround Conditions
∆EJab mean max min Std Yb1 Ybn LA1 LAn
M1 vs M2 0.17 0.20 0.02 0.04 19.7 20.96 13.33 13.97
M1 vs M3 1.94 3.30 1.15 0.58 19.7 18.6 13.33 13.06
M1 vs M4 0.19 0.23 0.02 0.05 19.7 18.91 13.33 12.60
M1 vs M5 0.08 0.10 0.01 0.02 19.7 19.52 13.33 13.01
M1 vs M6(1) 1.34 1.62 0.15 0.33 19.7 28.37 13.33 18.91
M1 vs M6(2) 2.90 3.86 0.33 0.87 19.7 6.58 13.33 4.39

3 .5

3
c o lo ur di ffe r enc e d E Ja mbm

2 .5

da rk
2
di m
1 .5
ave ra ge

0 .5

0
m e th o d 1/ 2 m e t ho d 1 /3 m et ho d 1 /4 m et h od 1 /5 m e th od 1 /6 m e th o d 1/ 6(2 )

Figure 4. Mean colour difference ∆EJab values in CIECAM02 J aM bM colour space of comparisons between Method1 and others respectively (under three surround
conditions).

Comparisons between Method 1 and the other methods were made surround area and image contents. More images and
for three different surround conditions (i.e. average, dim and dark). psychophysical experiments are required to evaluate these methods.
As shown in Figure 4, Method 4 (in which the captured image of
24-colour chart itself was treated as a complex stimulus from which It should be noted that all the results were obtained from one single
the mean value of the luminance channel Yb was derived) was found image  the digital 24-colour chart. The method (Method 4), which
to perform well with the smallest colour difference in CIECAM J, has the best agreement with the standard method (Method 1), uses
aM, bM colour space. Method 5 also performed well under dark the mean value of the luminance channel for the captured image as
conditions, but relatively worse than those under dim conditions, the luminance factor of background (Yb) which is dependent on
and even worse for average conditions. This might be due to the image contents. This method performs well with colour patches
fact that Method 5 is dependent on the luminance level for the under specific viewing conditions but there is still room for

13th Color Imaging Conference Final Program and Proceedings 73

investigation of complex images under complex viewing conditions 7. ISO 3664:2000, “Viewing conditions for graphic technology and
and the psychophysical experiment is ongoing. photography”.
8. P.J. Green, “Gamut Mapping and Appearance models in Graphic Arts
References Colour management”, PhD Thesis. University of Derby, UK (2003).
9. R.S. Berns, “Methods for Characterizing CRT displays”, Displays,
1. C. Fu and M.R. Luo, “Using a Digital Camera for Determining
16:173-182 (1996).
Viewing Parameters”, the 10th Congress of the International Colour
10. G. Hong, M.R. Luo and P.A. Rhodes, “A Study of Digital Camera
Association Conference, Granada, Spain, 1195-1198 (2005).
Colorimetric Characterisation Based on Polynomial Modelling”,
2. M.R. Luo and R.W.G. Hunt, “The structures of the CIE 1997 color
Color Res. App. 26: 76-84, (2001).
appearance model (CIECAM97s)”, Color Res. Appl., 23 138-146
11. CIE Technical Report, “Methods for Deriving Colour Differences in
(1998).
Images ”, CIE TC8-02 (Draft Version 0.7), 2004
3. CIE, Publication 131, “The CIE 1997 Interim Colour Appearance
12. M.D. Fairchild and G.M. Johnson, “Meet iCAM: A Next-Generation
Model (Simple Version) CIECAM97s”, CIE Central Bureau, Vienna,
Color Appearance Model”, IS&T/SID, 10th Color Imaging
Austria (1998).
Conference, Scottsdale, 33-38 (2002).
4. N. Moroney, M.D. Fairchild, R.W.G. Hunt, C. Li, M.R. Luo and T.
13. M.D. Fairchild and G. M. Johnson, “The iCAM Framework for
Newman, “The CIECAM02 Color Appearance Model”, The tenth
Image Appearance, Image Differences, and Image Quality”, Journal
Color Imaging Conference, IS&T and SID, Scottsdale, Arizona, 2002
of Electronic Imaging, 13 126-138 (2004).
23-27 (2002).
14. C. Li, M.R. Luo and G. Cui, “Colour Differences Evaluation Using
5. CIE Technical Report, “A Colour Appearance Model for Colour
Colour Appearance Models”, IS&T/SID Eleventh Color Imaging
Management Systems: CIECAM02”, CIE 159: 2004
Conference, 127-130 (2004).
6. R.W.G. Hunt, “Measuring Colour”, 3rd Edition, Fountain Press, U. K.
(1998).

74 Society for Imaging Science and Technology & Society for Information Display