Académique Documents
Professionnel Documents
Culture Documents
Abstract: This study investigates colour harmony in visual Ou9) defined colour harmony as an interrelationship of col-
experiments in order to develop a new quantitative colour ours. The main principles of these studies are ‘‘complemen-
harmony model. On the basis of new experimental results, tary’’ and ‘‘analogous’’ but these concepts are not consistent
colour harmony formulae were developed to predict colour among the studies. Also, the colour wheel10 was often
harmony from the CIECAM02 hue, chroma, and lightness adopted as a tool to define these basic relationships. Other
correlates of the members of two- or three-colour combina- authors (Judd and Wyszecki10) define colour harmony as a
tions. In the experiments, observers were presented two- more universal concept: ‘‘when two or more colours seen in
and three-colour combinations displayed on a well-charac- neighboring areas produce a pleasing effect, they are said
terized CRT monitor in a dark room. Colour harmony was to produce colour harmony.’’ Also, there is no consistency
estimated visually on an 11 category scale from 25 among the principles and the keywords of colour harmony.
(meaning completely disharmonious) to þ5 (meaning com- It is completeness according to Goethe,6 order according to
pletely harmonious), including 0 as the neutral colour har- Nemcsics5 and Chevreul,7 and balance according to Mun-
mony impression. From these results, mathematical models sell.2
of colour harmony were developed. The visual results were A quantitative model for two-colour combinations based
also compared with classical colour harmony theories. Two on the CIELAB colour space was developed by Ou et al.9
supplementary experiments were also carried out: one of We carried out computations11 to investigate and compare
them tested the main principles of colour harmony with real this model with classical colour harmony theories. Results
Munsell colour chips, and another one compared the visual showed that there was a significant difference between the
rating of the new models with existing colour harmony predictions of this model and classical harmony theories.
theories. Ó 2009 Wiley Periodicals, Inc. Col Res Appl, 35, 34 – 49, We also conducted a visual experiment11 to investigate the
2010; Published online 17 November 2009 in Wiley InterScience (www. correlation between Ou’s model and our visual dataset, but
interscience.wiley.com). DOI 10.1002/col.20558 only a weak correlation of r2 ¼ 0.30 was found. One possi-
ble reason of this phenomenon can be the different ethnic
Key words: colour harmony; colour harmony formula; origins of the observers, hence our observers were univer-
colour harmony rendering; colour rendering sity students from Hungary, while in Ou’s model, they were
Chinese observers.
INTRODUCTION
Nowadays, choosing of colours that are harmonious in
Colour theory has long had the goal of predicting or speci- a defined lighting situation is an important task not only
fying those colour combinations that would ‘‘work well to- for artists but also for engineers and for the overwhelming
gether’’ or appear harmonious. Scientists and artists of the majority of end users of colour products. It is important
last centuries (Itten,1 Munsell,2 Ostwald3) and nowadays in fashion design just as in developing programs for reha-
(Nemcsics4,5) developed colour order systems, where they bilitation12 and teaching13. If we want to describe the col-
defined rules to establish harmonic sets of colours. These our quality of light sources with a new harmony rendering
colour harmony studies were based on the orderly arrange- metric, correlating well with visual results, a new model
ment of colours in the colour order system. The second of colour harmony, based on a set of new visual experi-
group of authors (Goethe,6 Chevreul,7 Moon and Spencer,8 ments needs to be developed.
In the following sections, first the experiments per-
formed on a CRT monitor will be described. Based on the
*Correspondence to: Ferenc Szabó (e-mail: szafe@szafeonline.hu). results of this experiment, colour harmony models for the
different colour sample combinations will be developed.
V
C 2009 Wiley Periodicals, Inc. Then the model will be tested on an independent data set,
FIRST EXPERIMENT
Stimuli and Procedure
A flat-screen colour CRT monitor with 6500K white
point and 116 cd/m2 peak white luminance was used to
display all stimuli. It had good colour channel independ-
ence hence the additive phosphor matrix characterization
model14 could be used together with a polynomial CRT
tone characteristics model. The monitor was characterized
and calibrated using a well-calibrated Photo Research PR-
705 spectroradiometer. Verification measurements showed
that the accuracy was within 2% for the tri-stimulus values
using this monitor characterization model.
All observations took place in a dark room where the
colour monitor was the only light source. Three series of
observations were carried out within 4 months with nine
observers. They were five female and four male university
students familiar with computers and computer monitors.
Colour vision was tested by the Munsell-Farnsworth 100
Hue Test. The viewing distance was 60 cm, the viewing
area of the monitor was 32.5 cm 3 24.5 cm.
Test colour samples were displayed on a middle gray
background (J ¼ 50). The background covered a rectan-
gular area of 29 cm 3 20 cm. There was a white frame
around the screen boundary outside the background. The FIG. 1. (a, b) Experimental images on the CRT monitor.
size of each colour patch was 4.5 3 4.5 cm2. The colour Examples of two- and three-colour combinations.
patches were spaced next to each other, see Figs. 1(a) and
1(b). The task of the observer was to rate his/her colour
harmony impression on a scale from 25 to þ5, by click- were selected from eight hues (with CIECAM02 hue
ing on the corresponding numbered button in the experi- angles of 0, 45, 90, 135, 180, 225, 270, and 315 degrees)
mental computer program. Five corresponded to best har- and each hue consisted of 8 tones: vivid, pale, dull, dark,
mony and 25 corresponded to worst harmony and 0 cor- light grayish, grayish, dark grayish, and the ‘‘spectral’’
responded to neutral colour harmony impression. After (i.e., most saturated) colours. The five achromatic colours
each scaling, a middle-gray background with a white included the grayscale colours of CIECAM at J ¼ 15, 30,
frame was shown for 2 s between the slides to eliminate 60, 80, and 95. To examine the repeatability of the ob-
the after-image effect. The colour stimuli were selected server responses, the whole experiment was repeated.
from all parts of the CIECAM02 colour appearance
model. Three-colour Harmony Experiments
Three-colour combinations consisted of three square col-
Two-colour Harmony Experiment
our patches presented in a triangle position, having a com-
In the two-colour harmony experiment, there were mon vertex in the middle, see Fig. 1(b). Test patches of
2346 colour stimuli. Each stimulus consisted of two three-colour combinations were arranged similarly to the
squared colour patches presented side by side, in the mid- patches used in the two-colour harmony experiments with
dle of the gray background. These stimuli were generated the three squares having common edges. Two series of
from 69 colours, combining all colours from this set of experiments were carried out in case of three-colour com-
69 [(69 3 68) 7 2 ¼ 2346]. This set also contained 273 binations. One of them investigated monochrome three-
two-colour combinations of monochromatic harmonies colour combinations. The other experiment investigated
(containing colours which share the same hue). The 69 trichromatic three-colour combinations, which consists of
colours were selected systematically from the CIECAM02 three samples of different hues. The monochrome experi-
colour space containing five achromatic and 64 chromatic ment consisted of 3322 monochromatic three-colour com-
test samples. Positions of the colour stimuli in the CIE- binations. Test samples were selected from 12 hues (with
CAM ac–bc diagram can be seen in Fig. 2(a), CIECAM02 CIECAM02 hue angles of 0, 30, 60, 90, 120, 150, 180,
correlates can be seen in Table Ia. The chromatic colours 210, 240, 270, 300, and 330 degrees) and each hue con-
sisted of eight tones: vivid, pale, dull, dark, light grayish, reduce the number of stimuli, but have still enough stim-
grayish, dark grayish, and the ‘‘spectral’’ (i.e., most satu- uli for statistical calculations and regression. The number
rated) colours. The five achromatic colours included the of stimuli was reduced to 14,280.
grayscale colours of CIECAM at J ¼ 15, 30, 60, 80, and About 7140 stimuli were generated by using 54 colours
95. Monochromatic three-colour combinations were com- having moderate chroma (C ¼ 20) and three different
posed from these set of samples, selecting three of them lightness levels [J ¼ 30, 50, 80; see Fig. 2(c)] by combin-
which had the same hue. The colour stimuli of the mono- ing these colours within the set. Another 7140 stimuli
chromatic three-colour harmony experiment are shown in were generated by using 54 colours having medium light-
Fig. 2(b), CIECAM02 correlates can be seen in Table Ib. ness (J ¼ 50) and three different values of chroma [C ¼
To examine the repeatability of the observer responses, the 10, 20, 30; see Fig. 2(d)], by combining these colours
whole experiment with monochromatic three-colour com- within the set. Latter 54 colours contained five achromatic
binations was repeated. and 49 chromatic colour samples. The chromatic colours
Before starting three-colour harmony experiments, we were selected from 12 hues (CIECAM02 hue angle: 0,
had the results of the two-colour harmony experiments 30, 60, 90, 120, 150, 180, 210, 240, 270, 300, and 330
and our hypothesis was that colour harmony impression is degrees). The five achromatic colours included the gray-
influenced by differences and sums of chroma, lightness, scale colours of CIECAM J ¼ 15, 30, 60, 80, and 95.
and hue. Using the same sampling of the CIECAM02 Chromatic colours can be divided into ‘‘equal property sec-
appearance model as in case of the two-colour harmony tions’’ by their lightness and chroma values, see Figs. 2(c)
experiments (five levels of lightness, three levels of and (d), CIECAM02 correlates can be seen in Table Ic, d.
chroma, and 12 hues) produces more than 50,000 three- To examine the repeatability of observer responses, the
colour combinations, which is impossible to use in a vis- presentation of 2 3 2000 randomly chosen stimuli from
ual experiment. Some simplification had to be chosen to both sets was repeated once.
need separate mathematical models for monochrome har- visual colour harmony scores tend to increase if we con-
monies, see Sections ‘‘Colour harmony model for mono- sider not only one perceptual attribute but also other per-
chrome two-colour combinations’’ and ‘‘Colour harmony ceptual attributes or factors (e.g., the equal hue, nonequal
model for monochrome three-colour combinations.’’ lightness group has higher visual harmony score than the
Results indicated that small lightness differences had nonequal lightness group).
low visual colour harmony scores, see Figs. 3(b) and 6(b). The following models hypothesize that the sums and
As lightness difference increases, the harmony score also differences of the perceptual colour attributes (J, C, h) of
increases. Colour pairs having higher average lightness the constituting colours of the compositions influence the
were proved to be more harmonious than pairs of lower av- impression of colour harmony additively. Our descriptor
erage lightness, see Figs. 3(d) and 6(d). For chroma differ- formulae of harmony are assembled from the lightness
ences, a linear decrease of colour harmony can be observed difference, chroma difference, hue difference, hue, and
if chroma difference increases, see Figs. 3(c) and 6(c). the average lightness of the composition.
From these results, some general rules can be formu- Our aim is the accurate modeling of colour harmony in
lated for colour harmony. By depicting these results in a case of two- and three-colour combinations. To satisfy
colour space, general patterns of colour harmony can be this criteria, visual data were collected in case of the four
observed. Attributes of colour combinations producing models separately, and—instead of averaging—four dif-
harmony or disharmony can be defined. From the ferent (but similar) curves were fitted to visual results to
perceptual attributes of the constituent colours, the extent model the effect of the hues of the constituent colours in
of perceived harmony can be predicted, see the next Sec- case of the monochrome and dichrome two-colour combi-
tion. nations (see Fig. 3) and monochrome and trichrome three-
colour combinations (see Fig. 4). The mathematical dif-
ference between these hue preference curves and the
NEW MODELS OF COLOUR HARMONY
modeled hue effect in9 can be seen immediately. At the
Classical colour harmony theories most often formulate moment, we can present only colour harmony models that
general directives in terms of hue, chroma, and lightness are valid in case of Hungarian observers at the same age
and their differences to make harmonious compositions. (the similar facts can be concluded in,9 which model is
Appendix has a comparative description of the present valid only for young Chinese observers). One kind of hue
visual results and classical colour harmony theories. As preference effect in case of different ethnic groups of
can be seen from Table AI (Appendix), the present mean observers can be studied only during a collaborative work,
where,
Colour Harmony Models for Three-Colour Combinations The meaning of the subscript (n-m) in Eq. (14) is num-
If the aim is to model colour harmony not only for bering the constituent two-colour combinations i.e. (1-2),
two-colour combinations but also for three-colour combi- (2-3), and (3-1).
nations or even for four or five constituent colours then As can be seen, only a small correlation coefficient of r2 ¼
the following very exciting question arises. Is it possible 0.49 was achieved by this method (by optimizing the weight-
to use the above mathematical models of two-colour har- ing factors of Eq. (6) to achieve maximum correlation). Fig-
monies to describe three, four, or five-colour combina- ure 5 shows that a new model must be formulated for three-
tions? A straightforward solution is resolving a three-col- colour combinations to predict perceived colour harmony.
our combination into three two-colour combinations. Thus, authors constructed colour harmony models for three-
Before developing new mathematical models for three- colour combinations. The principle was the same as in case of
colour combinations, authors tried to model the present two-colour combinations: descriptor models for monochro-
three-colour combination experimental dataset with two- matic and tri-chromatic three-colour combinations were
colour harmony formulae. Latter two-colour combinations developed.
can be described by the formulae introduced earlier, see
Eqs. (1)–(12). Each three-colour combination used in the
Colour Harmony Model for Monochromatic
visual harmony experiments (see Section ‘‘Three-colour
Three-Colour Combinations
harmony experiment’’ earlier) was divided into three two-
colour combinations and colour harmony prediction was In case of monochromatic three-colour combinations,
made by averaging the values of the two colour harmony the hues of the three constituent colours are equal or
formulae. This method lead to a correlation coefficient of within the following limit: (|h1 2 h2| \ 58, |h1 2 h3| \ 58,
r2 ¼ 0.32. Reweighting the weights in Eq. (6) improved |h2 2 h3| \ 58). The lightness difference can be defined as
the method (r2 ¼ 0.49) and lead to Eqs. (13), (14) and to the sum of absolute lightness differences among any two of
the results depicted in Fig. 5. the constituents:
The chroma difference can be defined in the same way: DJ ¼ jJmid Jmin j þ jJmid Jmax j (25)
of these two-colour combinations (which were composed was eliminated (at the time of the ranking secondary sam-
from two Munsell chips) were made with the help of our ples for the primary 5PB 7/6 sample, the 5PB 5/6 second-
two-colour harmony formula. Viewing situation was con- ary sample was eliminated). So, observers were not
sidered as defining CIECAM02 parameters. allowed to form monochromatic harmonies.
For hue difference investigation, nine other Munsell The investigation of Munsell Value differences was
chips were selected with the same chroma and medium done using two procedures: The four primary samples
lightness difference as secondary samples, from 10 hues: were selected for investigation, and for each hue the Mun-
5PB 5/6, 5P 5/6, 5RP 5/6, 5R 5/6, 5YR 5/6, 5Y 5/6, 5GY sell Value levels 2/, 4/, 6/, 8/, 10/ were tested. In the first
5/6, 5G 5/6, 5BG 5/6, and 5B 5/6. Observers had two case the hue and chroma of the secondary samples (with
tasks. First, they had to choose for each of the four pri- Munsell value levels 2/, 4/, 6/, 8/, 10/) were the same as
mary samples the other member of the combination from the primary samples, so observers composed monochro-
the secondary set of nine chips to produce the most har- matic two-colour combinations. In the second case the
monious and the most disharmonious two-colour composi- chroma of the test samples was the same, but the hue dif-
tions. After that, they had to rank the other seven sam- ference was chosen to be different but optimal (neigh-
ples, and to order the nine two-colour combinations. boring hues with 368 of hue difference), so observers
Notice that all of the 10 main Munsell hues were between were asked to compose dichromatic two-colour combina-
the choices and therefore, observers could compose tions. Observers had the same task as in the case of the
monochromatic harmonies, too. hue difference tests.
By doing the investigation of hue difference, a second The investigation of chroma difference was carried out
part of observation was also carried out: in case of every by the aid of Munsell chips having equal hue and light-
primary sample, the secondary sample of the same hue ness. Four hues were selected for investigation, and for
each hue the Munsell Chroma levels /2, /4, /6, /8, /10 Comparison of Classical and Modern Theories
were tested. Observers had the same task, as in the case of Colour Harmony
of hue difference and lightness difference tests.
In our opinion, a verbal scale with elements: ‘‘best har- This section investigates and compares colour harmony
mony, second best harmony, third best harmony. . .’’ is theories and models, in order to reveal the theory which
much easier to use for the observer during this task as the produces the most harmonious colour composition for the
original scale from 25 to þ5. The expression ‘‘best har- observer. Beyond classical colour harmony theories,2,4,15
mony’’ was then transformed to maximum harmony score the predictive harmony model of Ou9 and our new
(which was the number of secondary samples used in the colour harmony formula for two-colour combinations are
experiment) and so on the ‘‘worst harmony’’ was trans- compared.
formed to minimum harmony score (þ1). To investigate colour harmony theories, a comparison
of the above mentioned colour harmony models and theo-
RESULTS
As can be seen in Fig. 9, by investigating hue differences,
the ‘‘similar hue’’ property was accentuated. Monochro-
matic harmonies or two-colour combinations with neighbor-
ing hues were found more harmonious than combinations
with large hue angle difference i.e., opposite hues.
Results showed that observers found the two-colour
combinations having different lightness more harmonious.
As can be seen from Fig. 10 if the hue angles are defined
to be the same and the two chroma levels are equal then
observers choose a sample with a medium lightness dif-
ference as the most harmonious. Samples with equal
lightness difference are the least harmonious.
The opposite result can be experienced in case of
chroma differences. If the hue is chosen to be the same, FIG. 11. Investigating chroma difference. Ranking of col-
and lightness difference is chosen to be optimal (three our combinations having equal hue and different lightness,
Munsell Value Difference) then the most preferred two- influence factor is: chroma difference. Scale values show
the place of the colour combination in a scale from 9 (best
colour combination can be arranged by minimal chroma harmony) to 1 (worst harmony) after averaging the ratings
differences [see section Models of Colour Harmony, Figs. of observers. Results are shown with 95% confidence
3(c) and 6(c)]. intervals.
Two-colour combinations
1/a Equal hue h1 ¼ h2 2.517 1.51 273
1/b Equal hue, equal lightness h1 ¼ h2 20.02 1.276 40
J1 ¼ J2
1/c Equal hue, nonequal lightness h1 ¼ h2 2.952 1.052 233
J1 = J2
1/d Equal hue, equal chroma h1 ¼ h2 3.495 0.575 58
C1 ¼ C2
1/e Equal hue, nonequal chroma h1 ¼ h2 2.253 1.577 215
C1 = C 2
2/a Neighboring hues |h12h2| ¼ 458 1.174 1.782 512
2/b Neighboring hues, equal lightness |h12h2| ¼ 458 0.34 1.628 160
J1 ¼ J2
2/c Neighboring hues, nonequal lightness |h12h2| ¼ 458 1.949 1.41 352
J1 = J2
3 Complementary hue |h12h2| ¼ 1808 20.22 1.51 256
3/a Complementary hue, equal lightness |h12h2| ¼ 1808 20.886 1.55 72
J1 ¼ J2
4 Equal lightness J1 ¼ J2 20.476 1.54 528
5 Nonequal lightness J1 = J2 1.91 1.48 1264
6 Equal lightness distance from medium lightness |J1 2 50| ¼ |J2 250| 2.19 1.31 693
7 Equal chroma C1 ¼ C2 3.26 1.06 568
8 Nonequal chroma C1 = C2 0.595 1.31 1224
Monochromatic three-colour combinations
1 Equal lightness J1 ¼ J2 ¼ J3 1.055 1.05 36
2 Nonequal lightness J1 = J2 = J3 2.23 1.11 2206
3 Equal chroma C1 ¼ C2 ¼ C3 3.305 1.18 34
4 Nonequal chroma C1 = C2 = C3 1.465 1.27 1608
Trichromatic three-colour combinations
1/a Neighboring hues |h12h2| ¼ 308 1.36 1.16 852
|h22h3| ¼ 308
1/b Neighboring hues, equal lightness |h12h2| ¼ 308 1.17 0.99 660
|h22h3| ¼ 308
J1 ¼ J2 ¼ J3
1/c Neighboring hues, nonequal lightness |h12h2| ¼ 308 1.42 1.37 504
|h22h3| ¼ 308
J1 = J2 = J3
1/d Neighboring hues, equal chroma |h12h2| ¼ 308 1.45 0.67 664
|h22h3| ¼ 308
C1 ¼ C2 ¼ C3
1/e Neighboring hues, nonequal chroma |h12h2| ¼ 308 1.37 1.39 514
|h22h3| ¼ 308
C1 = C2 = C3
2 Equal hue differences |h12h2| ¼ 1208 21 1.03 216
|h12h3| ¼ 1208
|h22h3| ¼ 1208
3 Equal lightness J1 ¼ J2 ¼ J3 0.02 1.23 7800
4 Nonequal lightness J1 = J2 = J3 0.08 1.49 1728
5 Equal chroma C1 ¼ C2 ¼ C3 2.14 2.04 7876
6 Nonequal chroma C1 = C2 = C3 20.18 1.32 2164
Hue difference is one of the most important and most produce a strong contrast or tension, because they com-
referred factors in classical colour harmony theories. plement each other when mixed, and they complete the
In case of two-colour combinations, equal hue and colour sense; others thought that the juxtapositions of
complementary hue are mentioned to be the most impor- complementary colours produce harmonious colour inter-
tant principles of creating harmony by many authors.2–8 actions.18 According to the present dataset, there is a sig-
The present visual results showed that observers found nificant contradiction between this principle and the pres-
the combination of two-colours harmonious if the member ent visual results, see Table AI and section New Models
colours of the combination shared the same hue. It can be of Colour Harmony, Figs. 4(a) and 7(a).
also seen that lightness and chroma differences [see Figs. The harmony of neighbouring hues is also frequently
3(b) and 3(c)] also have a relevant effect even if they are used by artist or colour designers.2–4 They tend to pro-
monochromatic harmonies. Some theorists and artists duce a single-hued or a dominant colour experience.
believed that juxtapositions of complementary colours These combinations produced also a pleasing effect in our