Affective Quantification of Color Gestalt

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 byguest on 25 September 2021

ABSTRACT ©2019 ISAST rules. To this point, color research on computer platforms computer on research color point, To this rules. orreconstruct color gestalt by managing color combination restore reliably cannot one methods, quantifiable Without ization dominatesstill their descriptions of color principles. fective factors for nearly two centuries, unquantified general af- studied haveresearchers color though even reason, this of the complicated interactions of their affective factors. For individual component parts. the only not background, the and foreground the in figure construct, e.g. they see the overall interrelations between the global a as together taken elements their of all with objects [2]. twentiethcentury of the stems from gestalt psychology of cognition of the early years taneous contrast and harmony, the term tween color gestalt and Chevreul’s color principles of simul- that determine how well colors the combine together [1]. terrelations of both color components and structural modes in subtle at hints generally design, multicolor of integrity can be challenging. The term plex patterns, and both creating and analyzing color designs Color design generally involves interlacing colors into com - and parametricdesign. researchoninteractivecolorcomputing and analyzable,invitingfurther Through thismethod,perceptionofcolorgestaltbecomescomputable color combinationthatstandsincontrasttoconventionalmethods. and userstudy. Theauthorproposesaholisticmethodforunderstanding models andinteractivetoolkitshebuiltforbothexperimentalpurposes preference andstructuralcomposition.Healsodescribesquantified factorsofharmony,quantification methodsformeasuringcolor-affective factors.Theauthorproposesasetof complex constructionsofaffective troubles researchersastheyseektounderstandandanalyzethe Integrated combinationofmultiplecolorsasitis,colorgestaltinvariably QuantificationofColorGestalt Affective l a c i n h c e T with thisissue. See www.mitpressjournals.org/toc/leon/52/2 forsupplementalfilesassociated Email: [email protected]. Guosheng Hu (artist, educator), China Academy of Art, Hangzhou, Zhejiang, China. It is difficult tomeasure and analyze color gestaltsbecause be relationship conceptual a clearly is there Although for Combinative Color Design Color Combinative for Factors Affective of Modeling o u G doi:10.1162/LEON_a_01197 A s r n e h t e l c i g Gestaltmeansthat humans see color gestalt, which refers to the

H u gestalt ashere used - - - multicolor combinations). in factors affective color studying when objects isolated as colors the take to researchers of tendency (the focus color single- combinationand color quantifying for method a of given disciplines in terms of tools, approaches, etc.), the lack terdisciplinary discrepancy (i.e. the differencesbetween two in gestalt: multicolor on research current to obstacles ing perplex three are there general, In relations.multicolor of factors affective to regard in especially infeasible, largely is • Interdisciplinary discrepancy. This discrepancy • conceptual systems, such as Goethe’s color and wheel simulation. This constitutes progress over previous pigment mixing, color lighting and computer color implements color quantification and exchange in systems. To some extent, color standardization values allow color exchange different between values specific of elements. The digitized element sell’s, Ostwald’s, CIE, etc., colors are formatted as standardizationColor . In color systems like Mun- mathematical mapping. artists and computer scientists inmetaphorical- This discrepancy is acommon problem forboth artists merely through synesthesia and metaphors. computers cannot interpret affective the feeling of gramming languages by used computer engineers— instinctivethe synesthesia of artists and pro the - crepancy to leads obstacles for interchange between that determine overall the color features. This dis- scientists may intuitively individual weigh the factors “vivid,” “elegant,” “gloomy” and on. so Computer ing metaphorical expressions, such as “sparkling,” appearances. may acolor Artists describe gestalt us- tend to deduce parameterized relations from color as awhole (color gestalt), computer while scientists artists are accustomed to grasping and judging colors neers define and analyze color. Generally speaking, to differencesthe ways in artists and computer engi- research and application. Such adiscrepancyis due computers when occurs are brought into color LEONARDO, Vol. 52,No.2, pp.157–163,2019 - - 157 Runge’s color sphere; however, these systems do little Modeling of Color Harmony Principles to quantify color-affective factors and expression ef- Color harmony is pleasing affective or aesthetic content fects. Standardized color systems made progress in that is produced by colors when combined together [6]. Re- color digitalization but did nothing to further under- searchers after Goethe proposed many ways to generalize standing and quantifying of color gestalt as integral the principles of color harmony. Ostwald insisted that color expression. harmony lies in ordered relations within elements, i.e. that • Single-color focus. In view of perceptual wholeness, the constitutive elements of harmonious colors are identical color gestalt lies in the interactions of each compo- to chromatic spans, such as gradation, isotones, isotints and nent in sight. In contrast, traditional color researchers so on [7,8]. Itten proposed color harmony principles based have usually taken colors as isolated objects, seldom on a series of graphic experiments, and he explained these perceiving color combinations as integral gestalts principles using the metaphor of musical harmony [9]. Moon and leaving interrelations of color components un- and Spencer divided the principles of harmony into catego- touched. Affective description and evaluation of color ries of similarity, ambiguity, contrast and glare, in accordance gestalt has remained a scarcely cultivated territory up with classical color harmony theory [10]. They also applied to this point. Meanwhile, the concept of color in an Birkhoff’s aesthetic measure M = O / C (where M stands for art context is more likely to refer to multicolor com- value of aesthetic measure; O for order as identity, similarity, bination than to single-color stimuli. Previous artistic contrast and area balance; and C stands for complexity as investigations under gestalt theory have been unani- summation of color amount and pairs of colors with element mously based on multicolor compositions [3–5]. In difference) [11]. Pieters constructed a similar measure based short, isolated colors make no sense in an art context on his own experimental results: Color Harmony = Satura- unless they are organized in certain interrelations that tion Harmony × (Hue Harmony + Lightness Harmony) [12]. form an integral object. Hård introduced a theoretical model for color harmony in which he divided harmony factors into three categories: color For these reasons, color researchers should seek methods interval, color chord and color tuning [13]. of affective factor quantification that can reconstruct and -ma However, all the aforementioned methods emerge from nipulate color gestalts. Each affective factor of color gestalt conceptual generalizations rather than computable mea- should be quantified and modeled for computing purposes. sures. Moreover, many existing methods are confined to However, it is not enough to merely quantify individual af- harmony principles of hue, leaving other element dimen- fective factors; composing interrelations between each factor sions—say, saturation or lightness—unaddressed. Well- is also essential. Though it is complicated and challenging to rounded harmony systems should be built on these all these do so, the integrity of different affective factors and the inter- factors together. Taking hue, saturation and lightness into relations of each should be considered, too. Another matter consideration, I built an integral model of harmony princi- is the diversity of individual observers. Although common ples by combining familial elements and rhythmic contrasts. sense exists in color perception, personal differences in re- Lenclos proposed the concept of familial colors to refer to sponse to color combinations can still be remarkable due those colors with identical elements such as hue, saturation to the observers’ personal instincts or cultural backgrounds. or lightness, because color combinations with corresponding To avoid the interference of these subtle factors, researchers elements tend to be harmonious [14]. In fact, Oswald had should make sure that each item under survey is confined already proposed a similar idea, but his work emphasized the in specific scope in order to isolate the affective factor from principles of contrasting elements, the opposite of familial irrelevant subjects. The abstract organization of affective fac- elements [15]. According to familial color theory, one can tors can then be restored and represented in a quantified set one or two of the three elements—i.e. hue, saturation or data model. lightness—to be identical for all the colors to make a color The main concept of my method consists of classifying combination harmonious. However, if all three elements are affective factors and constructing quantified models for each set in accordance with the familial rule, colors will become individual factor. I executed three research projects on as- totally identical. Colors can only be distinct from one an- pects of, respectively, harmony principles, color preferences other if at least one element that fulfills the contrasts has and composition factors. In the first project, I built a quan- different values. This leads to the other principle—rhythmic tified model of harmony principles for harmony-acquiring span for element contrast. and color-scheme generating. For preference quantification, Rhythmic contrast refers to the element value contrast I built a six-dimensional model of color preference traits to in repeated steps. Similar to the rhythm of music, rhythmic evaluate and reconstruct color combinations. In the project color contrast can be acquired by equal contrasts of element of composition factors modeling, I quantified the structural values. The idea of the color-rhythmic principle comes from factors and their relations for structural assessment and com- Ostwald’s concept of color order [16]. According to Jacob- position rebuilding. Although there are other digitized color son’s proposition, rhythm of color contrast can be acquired spaces such as CIE XYZ, CIE LAB, RGB cube and so on, via equidistant color distribution in Ostwald’s color space. I built my quantified models on HSL color space to make Conversely, unordered contrasts destroy rhythm and har- them more intuitive to artists, because artists are usually ac- mony, with some colors emphasized and others weakened. customed to the hue-saturation-lightness form. In accordance with the rhythmic principle, rhythmic span

158 Hu, Affective Quantification of Color Gestalt

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 by guest on 25 September 2021 to be purer and calmer. However, if all three elements follow the rhythmic contrast principle without familial dimension, harmony will be less likely, but the color combination may be more colorful and noisy. Following the two principles, harmonious color combinations can be constructed regardless of the diverse values of each component color. Fig. 1. The rhythmic principle of hue, saturation and lightness. Elements of all colors I have built a color scheme generation tool- are in identical steps that make the contrasts look rhythmic. (© Guosheng Hu) kit of the model for user study (Fig. 2). With this toolkit, a user can generate harmonious color combinations by choosing and adjust- ing harmonious structure style in the GUI. The originality of the harmonious structures lies in global color interrelations and quantified element valuables. Hence, features of color combination can be quantified as global color relations rather than as the summation of individual colors. Further information and user study data for the project has also been published [17].

Quantification of Preferential Traits Color preference is a kind of affective bias both for individual colors and for color combinations. However, existing research focuses mainly on individual color preferences; it seldom focuses on preferences for multicolor combinations. This may be due to the dif- Fig. 2. Color scheme generator based on the harmonious model. I designed the harmonious ficulties inherent in analyzing the complex color scheme generator for both user study and design application. To acquire harmonious color combinations, users need only set the number of colors and adjust the global color interrelations, constructions of color combinations with because the system follows the harmony principles. (© Guosheng Hu) multidimensional interrelation of elements. Nevertheless, there are still some contribu- tions in the research related to multicolor determines the contrast of color combination (Fig. 1). The combination. Chuang studied preference in two-color com- rhythmic structure of a color combination is constructed binations and discovered that, in the preferred combinations, with identical spans of one or two element dimensions that both components of color pairs were dominant, because do not follow the familial principle. people tended to favor those pairs composed of preferen- With familial and rhythmic principles combined, six styles tial colors [18]. Palmer and Griscom found that correlations of harmonious structure can be formed (Table 1). Those with between preference and harmony are reliably positive [19]. only one element following the familial principle, and the Schloss surveyed different opinions on preference, harmony other two elements in rhythmic contrast, tend to be more col- and similarity, and found remarkable effect of element re- orful and variable, while those with two elements following lations on two-color preference [20]. Ou et al. surveyed the familial principle, and only one in rhythmic contrast, tend perception influence on multicolor combinations, such as gender, culture and other affective factors [21]. TABLE 1. Six styles of harmonious structure with either one- or two- When it comes to preference research, it is crucial to be- uniform elements. gin by classifying the viewer subjects. It is meaningless to talk about color preference without considering viewers’ One-uniform element Two-uniform elements perception traits. However, few research projects have focused on individual preference, let alone further research in H , S , L H , S , L U R R R U U specific cases of various subjects. The reason may lie in two

HR, SU, LR HU, SR, LU aspects: (1) although preferences of viewers may be significantly different without obvious common traits, data from a H , S , L H , S , L R R U U U R large group of homogeneous subjects are essential for some purposes; and (2) individual data may be less important to HR, SR and LR represent hue, saturation and lightness elements in rhythmic

contrast; HU, SU and LU represent the uniform elements that follow the familial engineers, and people can seldom objectively describe their principle. (© Guosheng Hu)

Hu, Affective Quantification of Color Gestalt 159

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 by guest on 25 September 2021 own preferential traits, so that researchers in that field pay possess identical structural parameters (Color Plate C, top). less attention to individual subjects. This method demonstrates a new way of understanding and I recognized that it was necessary to find a feasible method analyzing multicolor combinations. The quantified prefer- for measuring personal preference traits of color combina- ence model can be applied to color preference analysis, as tion so that color preference could be considered on a per- well as to color research and application. sonal level. For this purpose, I designed a quantified model for measuring combinative color features of personal pref- Modeling of Composition Factors erence. This model allows preferential color features to be Assuming a set group of component colors, composition is determined and preferential color combinations to be re- also an influential factor for color gestalt appearance. Re- constructed. Preferential color combination lies in acquiring search on color composition stems mainly from the studies personal preference parameters via the quantified preferen- of De Stijl artists in the early 1900s, such as Mondrian’s paint- tial measure. ing experiments on composition of plastic units and colors The preferential measure consists of the color distribution [23]. At that time, color composition research witnessed great parameters in the HSL color space: the locations of the colors progress through intensive experiments and analysis at the and the covered spans of the colors. While the colors them- Bauhaus. Color composition theories appear abundantly in selves are determined by their own elements (hue, saturation the works and classes of teachers like Kandinsky, Klee and and lightness), location parameters define the global color Itten [24]. Itten insisted that composition units—such as tendencies of the color combination, and span parameters shape, area and block—have a significant influence on color determine the overall contrasts in each element dimension. presentation. Albers took a similar approach and made sub- For example, the higher the saturation span value is, the more stantial progress in visual experiments [25]. He explored intense the saturation contrast of a color combination will be. the interaction between color and composition through ex- Every color combination can be represented as value codes perimental painting and exposed the delicate compositional of the six-dimensional parameters shown in Fig. 3. Prefer- influence on pattern and color. These studies of color com- ential color combinations then can be constructed based on position theoretically originated from gestalt psychology. the personal values of these parameters. That is what I mean Much of the explorations of the De Stijl artists, the Bauhaus by “the quantified model for preferential traits”—the model school and Arnheim can be traced back to the experiments measures the specific viewer’s preferential traits in parameter of Wertheimer and Rubin, who emphasized the wholeness values. I built an interactive system on the model to access of formative arts and the interrelation of plastic units [26,27]. personal data about color preferences and to generate pref- However, these studies only revealed the influence of erential color combinations in reverse [22]. composition factors but are not adequate to describe that The preferential quantification method translates color influence in quantifiable terms. Another problem was that combination characteristics into certain structural color re- these studies usually focused on area proportions while color lations rather than a totality of individual component colors. gestalt is also influenced by other factors, such as position, It can precisely measure out features of color combination shape and the interrelations of these factors. regardless of number of colors and individual color values. In a survey of the structural factors, three factors—the In other words, color combination effects will be similar no area, distribution and patch amount of each component matter how many colors the color combinations have and color—along with their interrelations, stood out prominently no matter what colors they are, if the color combinations (Color Plate C, bottom). By borrowing ideas from Kansei engineering and affective computing domains, I was able to build a computable composition model accounting for these factors. Structural interrelations can be represented as quantified parameters via the model. Composition features and characteristics of each component color can also be evaluated. For the purpose of quantification, I designed three linear scales for measuring the structural factors. Colors of larger area usually play predominant roles in images, while the smaller areas of color appear less significant. Area Fig. 3. A four-color combination and its six-dimensional parameters. The six dimensions on the right are proportion of colors also has influ- the structural scales, with the range of values inside parentheses, which are the basis of the quantified model ence on figure-ground relationship. for preferential traits. The left image is composed as the certain values of the six-dimensional model on the To quantify the proportion in scale right. With the parameter values of the six-dimensional model, one can also tell the overall color features of the color combination. Thus, color combination preferences can be evaluated and presented by parameters value, I set the total area of an image of the six scales. (© Guosheng Hu) to 1, so the range of area proportion

160 Hu, Affective Quantification of Color Gestalt

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 by guest on 25 September 2021 for any given color is between 0 and 1. Color distribution plays an important role in color emphasis and figure-ground relationship. A centralized color is usually emphasized as the primary figure or dominant part, while a dispersive color usually fades into the background texture by cross-juxtapos- ing with other colors. To quantify the distribution of a color component, I set the distribution parameter between 0 and 1, using the distribution value 1 for a color that is absolutely dispersive and 0 for a color that is concentrated in a single area. To measure the distribution value of certain colors, I conducted a user study to help me build the algorithm. A color can be laid in a continuous area enclosed by other colors or it can be segmented into separate patches. This determines whether the shape of a color is a continuous shape or decentralized fragments. A lesser patch amount means the color is emphasized as a solid figure, while a greater patch amount means the color is interwoven with other colors. The Fig. 5. Visualization toolkit for color composition factors. The three-dimensional patch amount of a color depends on the image configuration frame (left) presents the composition feature of the pattern (upper right). itself. Finally, I count each continuous area of color to get the Percentage of area, patch amount and distribution (between 0, for totally total number of patches. concentrated, and 1, for extremely dispersed) are displayed for each color. Taking all three scales into consideration, each color of an The positions of the representative points present the structural features of the colors, and the size and shape of the polyhedron presents the overall relations image has certain parameter values, respectively, on the axis of the color image. The more dispersive the representative points are (that is, of area, patch amount and distribution. The character of each the larger the polyhedron is), the more separately characterized the colors in color is codetermined by these structural parameters. When the image. Meanwhile, the contrast between the foreground figure and the the comprehensive model is built at the conjunction of the background color will be more prominent. (© Guosheng Hu) three axes, the structural interrelations of the colors can be visualized, because every point in this coordinate space em- Although it involves only three structural factors, the af- bodies a certain set of parameters. Figure 4 is the coordinate fective evaluating cube presents a new method for under- cube built on the three dimensions of the compositional fac- standing and analyzing the composition of color gestalt. The tors. Composition features of both integral color image and quantified structural model can also make color composition each color can be visualized and analyzed in this coordinate more legible and analyzable. In the field of application, this space. The overall structural features can also be estimated model can be applied to color gestalt assessment, automatic according to the distribution of representative points in it. To pattern design, data mining for structural features and so on. test the coordinate model, I designed a visualization toolkit Further technical details are available elsewhere [28]. to evaluate color images (Fig. 5). Discussion My interest in affective quantification of color focuses on the interrelations of the colors and the integrity of the color combination, namely color gestalt, so that my method of color affective quantification is to evaluate the combinative color relations but not to measure certain values for individual colors. The affective feature of each color combination can be measured as certain parameter values in these models. Effects of color combination can be evaluated relatively within the affective models. These models and methods can be adopted in color computing and computer-aided design. And as the data structure of color gestalt is embodied, these models will benefit color data analysis in artificial intelligence systems. Considering interrelations of component colors, these models can also help further research on color gestalt. The three models of affective quantification were built separately on harmony, preference and composition factors. They take multicolor gestalt as an integral object by model- Fig. 4. The coordinate cube of three structural scales (area size, patch amount ing the affective factors rather than defining each component and distribution). Each point in the space has three parameters, and every color, as conventional methods do. These affective factors part of the coordinate space embodies a certain affective feature. When color image is evaluated in this cube, each component color can be represented and their interrelations respectively affect color appearance, as a certain point of the three-dimensional values. Structural features of both but that does not mean that color affect lies only in the three individual colors and their interrelations can be visualized. (© Guosheng Hu)

Hu, Affective Quantification of Color Gestalt 161

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 by guest on 25 September 2021 Fig. 6. Echoes Talking. A generative work with the methods of harmony modeling and preferential traits measuring involved, 2015. (© Guosheng Hu)

categories of factors. This article aims to introduce a quan- dimensions. A more practicable way to conduct color gestalt tification method for color appearance study and to inspire study would be to build a synthesized model based on more new methods for color gestalt research (Fig. 6). rounded achievement of color affective quantification. Thus, For technical purposes, color affective factors are clas- I plan to undertake further studies in two directions: ana- sified into individual items and then intensively surveyed lyzing and quantifying more affective factors and construct- one by one. However, it is still difficult to restore the com- ing an integral computing model of synthesized affective plicated color gestalts through isolated models of affective factors.

References 3 R. Arnheim, “Progress in Color Composition,” Leonardo 20, No. 2, 165–168 (1987). 1 A. Hård and L. Sivik, “Theory of Colors in Combination: A De- scriptive Model Related to the NCS Color-Order System,” COLOR 4 T.V. Doesburg, “Review of Hammer and Saw,” De Stijl 1, No. 3 (1917). Research and Application 26, No. 1, 4–28 (2001). 5 C.V. Campen, “Early Abstract Art and Experimental Gestalt Psy- chology,” Leonardo 30, No. 2, 133–136 (1997). 2 M.E. Chevreul, The Principles of Harmony and Contrast of Colors, and Their Applications to the Arts (1855), C. Martel, trans. (Whitefish, 6 K.E. Burchett, “Color Harmony,” COLOR Research and Application MT: Kessinger Publishing, 2009). 27, No. 1, 28–31 (2002).

162 Hu, Affective Quantification of Color Gestalt

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 by guest on 25 September 2021 7 W. Ostwald, Die Harmonie Der Farben (Leipzig: Verlag Unesma, Bibliography 1923). Fairchild, M.D. Color Appearance Models, 3rd Ed. (John Wiley & Sons, 8 E. Jacobson, The Color Harmony Manual and How to Use It (Chicago: 2013). Color Laboratories Division Container Corporation of America, 1942). Goethe, J.W.V. Theory of Colours, C.L. Eastlake, trans. (London: John Murray, 1840). 9 J. Itten, The Elements of Color (New York: Van Nostrand Reinhold, 1970). Healey, C.G., and Enns, J.T. “Attention and Visual Memory in Visualiza- tion and Computer Graphics,” IEEE Transactions on Visualization 10 P. Moon and D.E. Spencer, “Geometric Formulation of Classical and Computer Graphics 18, No. 7, 1170–1188 (2012). Color Harmony,” Journal of the Optical Society of America 34, No. 1, 46–50 (1944). Hsiao, S-W; Chiu, F-Y; and Hsu, H-Y. “A Computer-Assisted Colour Selection System Based on Aesthetic Measure for Colour Harmony 11 P. Moon and D.E. Spencer, “Aesthetic Measure Applied to Color Har- and Fuzzy Logic Theory,”COLOR Research and Application 33, No. 5, mony,” Journal of the Optical Society of America 34, No. 4, 234–242 411–423 (2008). (1944). 12 J.M. Pieters, “A Conjoint Measurement Approach to Color Har- Itten, J. The Art of Color: The Subjective Experience and Objective Ratio- mony,” Perception & Psychophysics 26, No. 4, 281–286 (1979). nale of Color (New York: Van Nostrand Reinhold, 1961). 13 Hård and Sivik [1]. Kandinsky, W., and Rebay, H. Point and Line to Plane (New York: Dover Publications, 1947). 14 J.P. Lenclos and D. Lenclos, Colors of the World: The Geography of Color (New York: W.W. Norton, 2004). Klee, P., and Faerna, J.M. Klee (New York: Cameo/Abrams, 1996). 15 Ostwald [7]. Kobayashi, S. Color Image Scale (Tokyo: Kodansha International Ltd., 1991). 16 Jacobson [8]. Land, E.H. “The Retinex Theory of Color Vision,” Scientific American 17 GS. Hu et al., “An Interactive Method for Generating Harmonious 237, No. 6, 108–128 (1977). Color Schemes,” COLOR Research and Application 39, No. 1, 70–78 (2014). Meier, B.; Spalter, A.; and Karelitz, D. “Interactive Color Palette Tools,” IEEE Computer Graphics and Applications 24, No. 3, 64–72 (2004). 18 M-C. Chuang and L-C. Ou, “Influence of a Holistic Color Interval on Color Harmony,” COLOR Research and Application 26, No. 1, 29–39 Moon, P., and Spencer, D.E. “Area in Color Harmony,” Journal of the (2001). Optical Society of America 34, No. 2, 93–103 (1944). 19 S.E. Palmer and W.S. Griscom, “Accounting for Taste: Individual Dif- Moretti, G.; Lyons, P.; and Marsland, S. “Computational Production of ferences in Preference for Harmony,” Psychonomic Bulletin & Review Colour Harmony, Part 1: A Prototype Colour Harmonization Tool,” 20, No. 3, 453–461 (2013). COLOR Research and Application 38, No. 3, 203–217 (2013). 20 K.B. Schloss and S.E. Palmer, “Aesthetic Response to Color Combi- Nemcsics, A. “Experimental Determination of Laws of Color Harmony. nations: Preference, Harmony, and Similarity,” Attention, Perception Part 3: Harmony Content of Different Hue Pairs,” COLOR Research & Psychophysics 73 (2011) pp. 551–571. and Application 34 (2009) pp. 33–44. 21 L-C. Ou et al., “A Study of Colour Emotion and Colour Preference. Part III: Colour Preference Modeling,” COLOR Research and Ap- Ostwald, W. The Colour Primer: A Basic Treatise on the Color System plication 29, No. 5, 381–389 (2004). of Wilhelm Ostwald (English edition by F. Birren) (New York: Van Nostrand Reinhold, 1969). 22 GS. Hu et al., “A User-Oriented Method for Preferential Color Scheme Generation,” COLOR Research and Application 40, No. 2, 147–156 (2015). Manuscript received 18 November 2014. 23 H. Friedel et al., Mondrian, De Stijl (München: Städtische Galerie im Lenbachhaus und Kunstbau; Ostfildern: Hatje Cantz, 2011). Guosheng Hu is an associate professor at China Academy 24 F. Whitford, The Bauhaus: Masters & Students by Themselves (New of Art. He received a PhD in computer engineering from Zhe- York: Overlook Press, 1993). jiang University and an MFA from China Academy of Art. His research interests are related to computer graphics, generative 25 J. Albers, Interaction of Color (new complete edition, New Haven and London: Yale Univ. Press, 2009). art and color research. 26 Arnheim [3]. 27 Campen [5]. 28 GS. Hu et al., “An Analytic Scale of Perceptual Measure for Color Composition,” COLOR Research and Application 41, No. 2, 165–174 (2016). Available at www.onlinelibrary.wiley.com/doi/10.1002 /col.21952/abstract.

Hu, Affective Quantification of Color Gestalt 163

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 by guest on 25 September 2021 Color Plate C: AFFECTIVE QUANTIFICATION OF COLOR GESTALT: COLOR PLATE C MODELING OF AFFECTIVE FACTORS FOR COMBINATIVE COLOR DESIGN

Top: Comparison of color combinations with identical preference parameters. Color combination of preferential traits can be restored via the structural measure of the six-dimensional model, regardless of the individual component colors. The left image has three colors, while the right one has five. All the colors of the two images are different. But the two images have similar color effects for the structural parameters they share. (© Guosheng Hu) Bottom: Comparison of compositional factors. Color gestalt is not only determined by colors themselves but is also significantly influenced by the composition they interweave. The images above of one color scheme are obviously different in proportion of area covered, distribution and patch amount of each component color. The three factors characterize the overall color effects. (© Guosheng Hu) (See article in this issue by Guosheng Hu.)

155

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/leon_a_01197 by guest on 25 September 2021