Expressive Animated Character Sequences UsingKnowledge-based Painterly Rendering

Hasti SeifiSchool of Interactive Arts and

Technology, Simon FraserUniversity

Surrey, BC, Canada V3T 0A3hsa43@sfu.ca

Steve DiPaolaSchool of Interactive Arts and

Technology, Simon FraserUniversity

250 -13450 102 AvenueSurrey, BC, Canada V3T 0A3

sdipaola@sfu.ca

Ali AryaCarleton University

Ottawa, ON, Canadaarya@carleton.ca

ABSTRACTIn this paper, we propose a technique to enhance emotionalexpressiveness in games and animations. Artists have usedcolors and painting techniques to convey emotions in theirpaintings for many years. Moreover, researchers have foundthat colors and line properties affect users’ emotions. Wepropose using painterly rendering for character sequencesin games and animations with a knowledge-based approach.This technique is especially useful for parametric facial se-quences. We introduce two parametric authoring tools foranimation and painterly rendering and a method to inte-grate them into a knowledge-based painterly rendering sys-tem. Furthermore, we present the results of a preliminarystudy on using this technique for facial expressions in stillimages. The results of the study show the effect of differ-ent color palettes on the intensity perceived for an emotionby users. The proposed technique can provide the animatorwith a depiction tool to enhance the emotional content of acharacter sequence in games and animations.

Categories and Subject DescriptorsI.3.7 [Computer Graphics]: Three-Dimensional Graphicsand Realism, Animation, Color;

Keywordsexpressive games, facial animation, Non-photorealistic Ren-dering, semantic painterly rendering, painterly animation,avatar, user perception

1. INTRODUCTIONThe expressiveness of character-based animations heavily re-lies on their success in conveying emotions to their viewers.In order to achieve high levels of expressiveness, today’sgame character sequences not only rely on the animationtechniques, but also on other elements such as lighting, andscene composition. The latter not only enhances the emo-

tional content of the scene, but also draws attention to thespecific regions designed for conveying the message. In asimilar way, painters use color, various brush strokes, andpainting techniques to emphasize certain features of a scenewhile leaving out unnecessary details. Edward Munch, andVincent Van Gogh are two exemplary artists that used col-ors to depict emotions in their work [4]. The use of texturefor inducing mood is also evident in different brush strokestyles used by artists [18].

The emotional effects of art and paintings on users point tothe importance of visual style on users’ perception and emo-tional responses. Some past psychological studies demon-strated that there are links between rendering style and one’sperception and feelings about a rendered object [8, 9, 12].Furthermore, some other studies investigated people’s asso-ciation between colors and emotion [12, 4, 19].

Painterly rendering, a subset of Non-photorealistic Render-ing (NPR) simulates the work of illustrators and portraitartists and is commonly reported as an expressive style [10,11, 14]. We propose using the expressive style of painterlyrendering to enhance the emotional content of facial char-acter sequences in games and animations. Specifically, wediscuss using semantic or knowledge-based painterly ren-dering systems whose painting process is informed by twosources: 1) Knowledge about the contents of the scene andits animation as a sequence. As our focus is mainly on fa-cial character sequences, thus facial features and emotionscompose the content of the sequence. In this regard, para-metric facial animation and available face standards makefacial animation very suitable for knowledge-based painterlyrendering. 2) Knowledge about the affective value of paint-ing parameters such as the color palettes and the shape ofbrush strokes. Such affective knowledge can be obtainedwith carefully-controlled user studies on face sequences. Apreliminary study in section 8 examines the affective valueof painterly rendering and colors on facial expressions.

The main contribution of this paper is proposing knowledge-based painterly rendering with two sources of informationand discussing its appropriateness for facial character se-quences. In addition, we detail two parameterized facial an-imation and painterly rendering systems and introduce ourmethod and extensions for integrating these two existingsystems into a knowledge-based painterly rendering system.

A knowledge-based technique applies color and painting pa-rameters depending on the content of the image and theeffect desired in different parts of the image. Thus, it canlead to conveying emotions to viewers more effectively andproviding more satisfactory emotional experience for them.Furthermore, using painterly rendering for facial charactersequences improves expressiveness with regard to the phe-nomenon of the uncanny valley of the human believability.While many games and animations try to achieve higher ex-pressiveness by adding to the reality of faces, this makes peo-ple notice even the slight flaws in the created faces. Movingto a completely synthetic painted world mitigates this is-sue and is less process-intensive compared to the algorithmsused for the highly realistic outputs.

2. RELATED WORKOur work is based on past psychological studies on the af-fective value of colors and line properties [19, 4, 12, 8].

Valdez and Mehrabian [19] found that brightness and satu-ration significantly affect emotions. They used the Pleasure-Arousal-Dominance (PAD) space for emotions [15] and sug-gested linear relationships between the Hue, Saturation, andBrightness components and the PAD dimensions. Accordingto their results, color brightness has a strong positive corre-lation to the pleasure dimension and the saturation compo-nent is correlated with the arousal axis in the PAD space.

Recently, da Pos and Green-Armytage [4] asked Europeanand Australian designers to choose first, a combination ofthree colors and then a single color that best fits each of thesix basic facial expressions. They found that there is an over-all consensus among designers in their color-emotion associ-ation. For instance, a great percentage of the participantsused red and black for anger; for surprise and happiness,they mostly chose yellow and orange hues. Colors for fearand sadness were mostly desaturated. This study suggeststhat there might be specific colors congruent to each emo-tion. We used findings of this study to design color palettesfor our preliminary study.

Some other studies demonstrated that various line styles cancommunicate affective messages. A study by Henver [12] ex-plored the affective value of colors and different shapes usedin a painting. According to their results, blue is expressiveof dignity, sadness, tenderness, and red is indicative of hap-piness, restlessness and agitation. Moreover, they realizedthat curves are tenderer and more sentimental compared tothe straight lines that are sad, serious, vigorous, and robust.These findings point to the importance of the visual style onviewers’ perception.

Duke et al. [8] investigated the affective qualities of NPRimages to provide suggestions for computer graphics. Theyconducted a series of experiments to show that different ele-ments of NPR illustrations could affect users’ feelings. Someof the results of their experiments show that: 1) NPR can beused to induce perception of safety; people associated jaggedoutlines to danger whereas they perceived straight outlinesas safe, 2) NPR can affect interpersonal relationships; inanother experiment, people associated line thickness, andcontinuity to a character’s strength, and 3) NPR can influ-ence both navigation and exploration behaviours; subjects in

their experiments tended to choose paths with higher levelsof visual details. They concluded that psychological theoriesand studies can be used to enhance the outputs of NPR ap-plications; and moreover psychology and NPR can mutuallybenefit from such interaction.

Furthermore, recent studies in human vision demonstratedthat texture properties of a painting can guide viewer’s gazethrough a portrait painting. DiPaola et al. [7] conducted aseries of eye tracking studies and demonstrated that Rem-brandt used lost and found edge techniques and a center ofinterest area with smaller brush sizes to guide users’ gazein a portrait painting. Our system is partially based onthis work and the findings of da Pos and Green-Armytage’sstudy [4].

3. OUR PROPOSED APPROACH: KNOWLEDGE-BASED PAINTERLY RENDERING

Having a high-level knowledge of a scene and its contentsis critical for an artist to make a concrete painting with astrong unified message. In a similar way, having knowledgeabout the contents of an input image in a painting algorithmallows the painting system to adjust painting parametersbased on the characteristics of the objects to be depictedtowards a meaningful output.

This type of NPR techniques is called semantic or content-based. Semantic NPR techniques have been used in a fewNPR systems. For example, Santella and DeCarlo [17] useddata from an eye-tracker to determine the regions of highersaliency in an image in order to paint those regions withmore details. O’Regan and Kokaram [16] used skin andedge detection techniques to recognize people’s head withina video sequence and set the appropriate painting parame-ters for those areas.

We propose semantic or knowledge-based painterly render-ing systems with two sources of information: 1) Knowledgeabout the contents of the scene and its animation as a se-quence. 2) Knowledge about the affective value of paintingparameters such as the color palettes and the shape of brushstrokes.

Regarding the first source of information, a semantic painterlyrendering system for sequences of faces requires knowledgeabout different parts of the face and the facial emotions ex-pressed. Meanwhile, the parametric approach to faces andfacial animation segments the face to meaningful, config-urable units. The knowledge of these facial parameters iswhat the proposed painterly system requires. Thus, we be-lieve facial sequences have great potential for knowledge-based rendering approach compared to the sequences of othercontents such as landscapes.

There are two well-structured parametric systems for faceand facial animation: 1) Ekman’s Facial Action Coding Sys-tem (FACS) and 2) the MPEG-4 standard for faces. FACSwas developed by Paul Ekman in the field of psychologyto study the human face and its emotions. MPEG-4 isanother standard commonly used by facial animation sys-tems. The MPEG-4 standard includes Facial Definition Pa-rameters (FDPs) and Facial Animation Parameters (FAPs).These two standards enable modular modification of facial

sequences. Moreover, they enable transmission of the fa-cial sequences between different systems and even from onehead model to another one. Thus, these standards allow usto communicate information about facial features and theirmovements in a sequence to a depiction system, which inturn allows more sophistication in the painting process andbetter visual outputs.

Moreover, we propose an additional layer (the second sourceof information) to the existing semantic systems which uti-lizes the knowledge about the affective value of painting pa-rameters, in order to make more expressive painterly out-puts. The repository of such affective knowledge needs tobe built by conducting controlled user studies. In section 8,we present the results of a user study which examines theeffect of color palette. The color palette is an important pa-rameter in a painting; the appropriate choice of the colorscan strengthen the emotional content of the painting (e.g.the color palette in scream by Edward Munch). Knowledgeabout the emotional impact of color palettes on facial expres-sions helps a painterly system to choose the colors based onthe desired effect. Similar studies on other painting param-eters such as the size and shape of brush strokes enhancesthe output of a painterly system.

As mentioned above, parametric authoring tools both in theanimation and depiction side help in better control over thepainting process and lead to more appealing results. In thenext section we describe a parameterized facial animationauthoring tool and a parameterized painterly rendering sys-tem proposed by past researchers [6, 5] and a method tointegrate them into a knowledge-based painterly animationsystem.

4. A PARAMETERIZED FACIAL ANIMA-TION AUTHORING TOOL

iFace is a parameterized facial animation authoring tool,which was developed as a set of .NET components in C# [6].iFace is based on the definition of Face Multimedia Object(FMO) which encapsulates all the functionality and datarelated to facial actions. It simplifies and streamlines thecreation of facial animations and provides programming in-terfaces and authoring tools for defining a face object, itsbehaviours, and animating it in static or interactive situ-ations. Four spaces of geometry, mood, personality, andknowledge in the framework cover the appearance and thebehavior of a face object.

Geometry: This parameter space is concerned with thephysical appearance and movements of facial points. It isgrouped into three levels of increasing abstraction: vertices,facial features/parameters, and facial regions. A user canmanipulate the face on all three levels. Higher levels in thehierarchy have access to the lower levels internally. Thisarchitecture helps in hiding the details from users and pro-grammers.

Mood: This space is concerned with short-term emotionsof a character and modifies how an action is animated.

Personality: This space deals with the long-term individ-ual characteristics and is mainly based on the facial cuesassociated with a given personality type. The association

between facial cues and personality types was derived fromthe past user studies [1].

Knowledge: This space is concerned with behavioural rules,stimulus-response association, and the required actions. Thisspace uses an XML-based language called FML. FML is aStructured Content Description mechanism, specifically de-signed for face animation. An FML file can incorporate asequence of actions by a head model or a set of paralleland event-based actions, which can be played back in iFace.Thus, FML provides iFace with a built-in functionality forcommunicating the emotional content of a facial sequence.

Thus, FML is useful for:

• Hierarchical representation of face animation from sim-ple moves to stories

• Timeline definition of the relation between facial ac-tions and external events. This includes parallel andsequential actions, and the choice of one action from aset based on an external event.

In addition, iFace provides a key frame editor toolkit forcreating and manipulating a key frame facial animation se-quence based on MPEG-4 facial parameters.

5. A PARAMETERIZED SEMANTIC DEPIC-TION SYSTEM

Figure 1: The process chart for making the Painterlysystem (from [6]).

For an NPR system to be flexible and to generate a wide va-riety of outputs and styles, it should provide the user with aset of controls. In such a system, the user or algorithm canadjust the parameters based on the input image and the de-sired effect/style. The ”Painterly”system by DiPaola [5], is aparameterized NPR toolkit developed for portraiture paint-ings. It takes advantage of portrait and facial knowledge inthe NPR process.

The Painterly program has been developed based on the col-lected qualitative data from art books and interviews withportrait painters. The qualitative data about how portraitartists achieve the final painting was then translated into aparametric model and incorporated into the painting algo-rithm (see Figure 1).

The Painterly system can differentiate and recognize areas ofcharacter scene, thereby approximating the cognitive knowl-

edge of a human painter. This functionality enables a tech-nical user to produce sophisticated paintings by applyingdifferent painting techniques on different parts of an inputimage.

The parameters in the Painterly fall into functional types:1) constant parameters such as brush size, 2) method pa-rameters such as the type of palette to use, and 3) processmethod parameters which guide the process flow of the otherparameter types.

The Painterly system receives the set of painting parametersin an XML format together with the name of the input imageand a number of matte files. Then, the system renders theinput image in an oil-painting style (see Figure 2).

Figure 2: Sample output rendered by the Painterlysystem using a source image and a painterly config-uration file

6. INTEGRATION OF THE ANIMATION ANDTHE PAINTERLY SYSTEMS

In this section, we explain how an exemplary parameterizedanimation system such as iFace can be used along with aparameterized painterly system (e.g. Painterly) to generatean NPR output.

The animator creates a facial sequence by setting appropri-ate values for the facial parameters in the key frame editor ofiFace.The resulting keyframe file from this step contains val-ues for facial parameters such as eye and mouth parametersover time. In a similar way, the animator sets emotion pa-rameters over the timeline of the sequence using iFace withthe new emotion based extensions to our FML language.The emotion information with timing is saved in an FMLfile. Figure 3 shows a sample FML which describes the sur-prise emotion. According to the file, the emotion starts atframe 20 and ends at frame 48. The ”timing” tag providesinformation about the onset, duration, and the offset of theemotion as well as the highest intensity that the emotionreaches on a scale of 0-100.

<fml>

<model><event name="kbd"/></model>

<story>

<emotion>

<hdmv type="sadness" begin="20" end="48"/>

<timing onset="8" duration="12"

offset="8" maxvalue="80"/>

<paint begin="20" end="48" csv="sadness.csv"/>

</emotion>

</story>

</fml>

Figure 3: FML

The Painterly rendering program for animations reads theinformation in the keyframe file and the FML file. Then, thePainterly program sets the appropriate painterly parametersto intensify the surprise emotion for frames 20 to 48. Know-ing the onset, duration and offset frames for the emotion,the Painterly program can change the painterly parametersfor surprise gradually over the sequence of frames to achievea more artistic output. As shown in Figure 3, the anima-tor can optionally specify the configuration (.csv) file for thepainterly parameters. If they do so, the specified frames arepainted using the parameters in the configuration file. Fig-ure 4 shows a diagram on the integration of the animationand painterly systems.

Figure 4: Integration diagram for painterly render-ing based on the emotion information.

In this method, the animator determines the curve for emo-tions over time in iFace, which then will be used by thedepiction system for painting the sequence. Furthermore, ifthe user wants to adjust the lower-level parameters in thedepiction system, he/she can draw the curve for the valuesof any desired painting parameter such as the brush size overtime. Then, the Painterly system will adjust other param-

eters based on the provided information and the emotionalcontent of the frames.

7. SAMPLE IMAGES WITH PAINTERLY EF-FECTS

Figure 5: Five color palettes for basic emotions.

We designed five color palettes based on color combinationsprovided in da Pos and Green-Armytage’s paper [4]. Threeof the color palettes were designed to enhance the four basicemotions of Anger, Fear, Joy and Surprise (see Figure 5).According to the results of that study, colors for fear aremainly desaturated while the designers’ choice of colors foranger were mainly red and black. Therefore, we designed thedark-red color palette for the anger emotion and the cool-light color palette for the fear emotions. Also, the resultof da Pos and Green-Armytage’s study [4] suggests similarcolor choices for joy and surprise by the designers. Thus,we designed one color palette for these two emotions, theyellow-orange color palette. The two other color paletteswere added for the purpose of our user study. Cool-darkcolor palette was added to analyze the effect of brightnessin the data. Finally, the natural color palette uses the samecolors as the original 3D frame and is considered as a baselinefor our analysis. Figure 5 shows the color palettes with thecorresponding emotions as captions.

Figure 6 shows one frame with the Computer Graphics stylein the middle and two painterly renderings of that image indifferent color and brush types in both sides. The left imageconveys sadness while the right image demonstrates a moreactive emotion; a combination of pain and anger.

8. PRELIMINARY STUDY AND APPLICA-TIONS

Figure 6: One image with two different painterlyeffects

A knowledge-based painterly system for faces requires knowl-edge about the emotional effect of painterly parameters suchas color palette and brush strokes applied in the context offace and facial expressions. Such knowledge is obtained bycareful user studies.

Thus, here we describe a preliminary study that we con-ducted on the effect of color palette and painterly renderingon facial expressions. The question of interest is whetherdifferent color palettes can impact the emotional intensityperceived for a given expression or not. Although humanscan express a wide range of facial emotions, six facial expres-sions are known to be universal. Also, there are well-definedfacial configuration for these expressions. In addition, manyother emotions are a combination of these basic emotions.Our study examines four of these six universal emotions:Anger, Joy, Fear, and Surprise.

Materials for the study were still images from animated fa-cial sequences of the four mentioned universal expressionsin medium and high intensity levels. The image for eachof the facial expressions was painted by the Painterly pro-gram in all the five color palettes (see Figure 5), totalling to40 images (4*5*2=40). We call the color palette designedfor an emotion the congruent palette for that emotion. Allother color palettes are considered incongruent for that emo-tion. 15 university students (6 female, 9 male, between 20-35years old) from various cultural backgrounds (Asian, Mid-dle East, and North America) participated in a 30 minutestesting session. During the session, the participants watchedall 40 images for three times (in random order) and chosethe intensity perceived for each facial expression on a Likertscale of 0 to 6 (zero being the neutral position and six thehighest intensity for the emotion).

Analysis of the results with a within-subject ANOVA testwith three factors of emotion type (4 basic emotions), con-gruency (3 levels: congruent, incongruent, natural), and in-tensity level (medium and high) yielded a main effect ofcongruency (F(2,28) = 8.86, p = .001). The results of Bon-ferroni’s post-hoc test showed that for all four emotions,the mean rating for congruent color palette to each emotion(M=3.74, SD=0.19) was significantly higher than the meanintensity rating for the incongruent color palettes (M=3.24,SD=0.18). This result indicates that an appropriate colorpalette increases the intensity of the emotions perceived bythe viewer. These preliminary results support the idea ofusing painterly rendering to affect the emotional content offacial expressions. Also confirmed is the importance of color

palettes as substantial tool for emotional depiction. Theseresults also provide information about the affective value ofdifferent color palettes for facial expressions.

The animation and movie industries usually use a varietyof techniques such as lighting and music to boost or flavorthe emotional impact of a scene. An exemplary feature of asoundtrack, which corresponds to emotions, is tempo. Mu-sic with fast tempo induces excitement while a slow tempoconveys calmness or sadness [13]. Movie and animation se-quences usually require carefully composed soundtracks toinvolve viewers in the story effectively [3]. Lighting is an-other effective tool for adding meaning and emotion to char-acters and scenes [20]. The number, colour, direction, andthe quality of a scene’s lightings has a direct impact on howthe characters and scenes are perceived and consequently onthe viewers’ emotional gain from the movie or animation se-quence. As an example, well-lighted scenes induce a pleasantmood while dark scenes are usually associated with sadnessand mystery [2]. These techniques have been used for manyyears as effective tools for adding to the emotional impactof movie and animation sequences. We believe that the pro-posed technique provides the animators with an additionaltool to intensify the emotional content of a scene.

Rapidly generated character sequences in games and com-mercials use different algorithms and artificial intelligencetechniques for adding expressiveness to facial and body move-ments in those sequences. The proposed technique in thispaper requires very little manual work by the game designerand is an efficient way for achieving better visual results infacial character sequences. In addition, the proposed tech-nique modifies the visual aspect of the scene in the final stageof the production. Thus, it can accompany almost any othermethod to achieve higher quality for the animated sequence.

Lastly, having knowledge about the affective content of theNPR elements not only has advantages for the gaming andanimation industry and any other application areas of theNPR systems, but also provides psychologists with a betterpicture of the human cognitive processes [9].

9. CONCLUSIONIn this paper, we propose a knowledge-based painterly ren-dering technique to enhance emotional expressiveness of gameand animation character sequences. This approach is espe-cially useful for facial sequences because of the well-structuredstandards for faces. We detail our parameterized anima-tion and painterly authoring tools and our extensions andmethod to integrate them. The painterly system receivesinformation about positions of facial features and the emo-tional content of the sequence from the animation system.Having this information, the system utilizes its affectiveknowledge of different lines and colors to apply painterlyeffects depending on the content of the sequence. Our pre-liminary study shows that such informed painterly renderingincreases the intensity perceived for facial expressions in stillimages. We continue our studies on the effect of color onanimated character sequences (dynamic stimuli). One an-ticipated direction for future research is using a continuousmeasurement for capturing users’ responses. For instance,using a continuous emotion space such as the PAD spacecould lead to capturing slight variations in perceived emo-

tion type and quality caused by changing colors. Examin-ing the effect of different painterly textures by using variousbrushes is another possible direction for future studies. Sim-ilar types of studies on the effect of painterly rendering for3D characters and avatars can provide better understandingof affective response to visual stimuli and thus enhance theproposed knowledge-based approach.

10. REFERENCES[1] A. Arya, L. N. Jefferies, J. T. Enns, and S. DiPaola.

Facial actions as visual cues for personality. ComputerAnimation and Virtual Worlds, 17(3-4):371–382, 2006.

[2] J. Birn. Digital lighting and rendering. New RidersPublishing Thousand Oaks, CA, USA, 2005.

[3] A. J. Cohen. Music as a source of emotion in film.Music and emotion: Theory and research, ed. PNJuslin & JA Sloboda, pages 249–72, 2001.

[4] O. da Pos and P. Green-Armytage. Facial expressions,colours and basic emotions. Colour: Design &Creativity, 1(1):2, 2007.

[5] S. DiPaola. Exploring a parameterised portraitpainting space. International Journal of Arts andTechnology, 2(1):82–93, 2009.

[6] S. DiPaola and C. Akai. Blending science knowledgeand AI gaming techniques for experiential learning. InProceedings of CGSA 2006 Symposium. YorkUniversity, 2006.

[7] S. DiPaola, C. Riebe, and J. T. Enns. Rembrandt’stextural agency: A shared perspective in visual artand science. Leonardo, 43(2):145–151, Apr. 2010.

[8] D. J. Duke, P. J. Barnard, N. Halper, and M. Mellin.Rendering and affect. In Computer Graphics Forum,volume 22, pages 359–368, 2003.

[9] N. Halper, M. Mellin, C. S. Herrmann, V. Linneweber,and T. Strothotte. Psychology and Non-Photorealisticrendering: The beginning of a beautiful relationship.Mensch and Computer 2003, Interaktion in Bewegung,2003.

[10] J. Hays and I. Essa. Image and video based painterlyanimation. In Proceedings of the 3rd internationalsymposium on Non-photorealistic animation andrendering, pages 113–120, 2004.

[11] A. Hertzmann. Algorithms for rendering in artisticstyles. PhD thesis, New York University, 2001.

[12] K. Hevner. Experimental studies of the affective valueof colors and lines. Journal of Applied Psychology,19(4):385–398, 1935.

[13] P. N. Juslin and J. A. Sloboda. Music and emotion:Theory and research. Oxford University Press Oxford,England, 2001.

[14] J. Lansdown and S. Schofield. Expressive rendering: Areview of nonphotorealistic techniques. IEEEComputer Graphics and Applications, pages 29–37,1995.

[15] A. Mehrabian. Pleasure-arousal-dominance: A generalframework for describing and measuring individualdifferences in temperament. Current Psychology,14(4):261–292, 1996.

[16] D. O’Regan and A. C. Kokaram. Skin-Awarestylization of video portraits. In 2009 Conference forVisual Media Production, pages 35–44, 2009.

[17] A. Santella and D. DeCarlo. Visual interest and NPR:an evaluation and manifesto. In Proceedings of the 3rdinternational symposium on Non-photorealisticanimation and rendering, page 150, 2004.

[18] M. Shugrina, M. Betke, and J. Collomosse. Empathicpainting: interactive stylization through observedemotional state. In Proceedings of the 4thinternational symposium on Non-photorealisticanimation and rendering, pages 87–96, 2006.

[19] P. Valdez and A. Mehrabian. Effects of color onemotions. Journal of Experimental Psychology:General, 123(4):394–409, 1994.