Performance of CIE94 for
Nonreference Conditions
Lewis D. Griffin,* Arsalan Sepehri
Medical Imaging Science IRG, King’s College, London SE1 9RT, United Kingdom
Received 1 December 2000; accepted 20 May 2001
Abstract: The CIE94 colour difference equations are known
to give good predictions of subjective colour differences for
viewing under reference conditions. We have psychophysi-
cally tested whether the equations also give useful predic-
tions when the conditions are relaxed. In particular, we
collected subjective colour difference data with (i) chro-
matic illuminants, (ii) textured samples, and (iii) large
colour differences. The predicted differences, to be com-
pared with the measured subjective differences, were cal-
culated by using a von Kries transform to predict colour
appearance under D65, followed by application of the
CIE94 equations (with the default lightness, chroma, and
hue weightings). We found that subjective colour differ-
ences were not predicted properly. However, if texture is
accounted for by modifying predicted colour differences by
a sample-pair dependent linear function, then prediction is
successful for all data points apart from those correspond-
ing to illumination under a strongly saturated red illumi-
nant. The fit of this model and its utility in predicting
illuminants that enhance colour differences was confirmed
by statistical tests. © 2002 Wiley Periodicals, Inc. Col Res Appl, 27,
108 –115, 2002; DOI 10.1002/col.10029
Key words: color difference equations; CIE94; colored
illuminants; chronic adaptation; color difference metrics
INTRODUCTION
Several researchers
1,2
have described methods for selecting
illuminants that boost particular colour differences. The
selection methods are based on calculations of predicted
colour differences using various colour distance formulas.
Although the use of such formulas for this purpose is
intuitively reasonable, their adequacy for such a task has not
been evaluated.
We report here the results of an experiment to test
whether a particular method of predicting colour differences
(von Kries followed by CIE94) is useful for such compu-
tations. Because the ultimate aim is practical, we examined
conditions that are more realistic than the standardized
conditions in which CIE94 has been validated. In particular,
we considered coloured illuminants, textured samples, and
subjective colour differences of large magnitude.*
As background, the remainder of the introduction reviews
(i) psychophysical measurement of colour differences, (ii)
colour metric formulas that aim to predict colour differ-
ences, (iii) chromatic adaptation transforms that aim to
discount the effect of chromatic illuminants, and (iv) illu-
minants optimized with respect to colour differences.
Measuring Subjective Colour Differences
A variety of psychophysical methods have been devel-
oped for measuring colour differences. These include: (i)
assignment of a numerical value to the difference or ratio
between the colour differences of two pairs of samples
3–5
;
(ii) choosing the most similar pair out of three
6
or more
7–8
samples; (iii) measuring response times in same/different
9
and search
10 –11
tasks; (iv) comparing colour differences by
seeing which best supports spontaneous grouping
12
; and (v)
adjusting a gray sample of variable lightness so that its
colour difference from a fixed gray equals in magnitude the
colour difference of two samples.
13–16
Several researchers have investigated how the details of
stimulus presentation affect colour difference. At least three
aspects have been shown to have an effect: (i) the back-
ground colour,
17
(ii) whether the samples are adjacent,
7,18
and (iii) the precise instructions given to the subjects.
19
Colour Metrics
Colour metrics are systems of equations that predict colour
differences. Their development has a long history that has been
*Correspondence to: Lewis D. Griffin (e-mail: Lewis.Griffin@kcl.ac.uk)
© 2002 Wiley Periodicals, Inc.
*The use of the terms subjective and magnitude mostly will be sup-
pressed in the remainder of the article.
108 COLOR research and application