Monica S. Castelhano and Kevin G. MunhallQueen’s University
An ongoing challenge in scene perception is identifying the factors that influence how we explore our visualworld. By using multiple versions of paintings as a tool to control for high-level influences, we show thatvariation in the visual details of a painting causes differences in observers’ gaze despite constant task andcontent. Further, we show that by switching locations of highly salient regions through textural manipulation,a corresponding switch in eye movement patterns is observed. Our results present the finding that salientregions and gaze behavior are not simply correlated; variation in saliency through textural differences causesan observer to direct their viewing accordingly. This work demonstrates the direct contribution of low-levelfactors in visual exploration by showing that examination of a scene, even for aesthetic purposes, can be easilymanipulated by altering the low-level properties and hence, the saliency of the scene.
When we look at art in a museum, we find ourselves in what thewriter Isaac Bashevis Singer called the chasm between the artist’sinner vision and its ultimate expression. Our work here shows thatthe fine details of the artist’s expression can guide visual explo-ration in this aesthetic task. Paintings and their visual examinationthus may reveal the operating principles of the human visualsystem.
It is well known that the human visual system is attracted tosome properties over others. Salient visual features such as localdifferences in contrast, orientation, color, and texture have shownto collectively predict gaze in natural scenes (Itti & Koch, 2001).In fact, the early primate visual system presents a common cross-species response to these visually prominent regions through atopographically coded saliency representation or map. Here weutilize the Itti and Koch model of saliency to analyze differencesin images. This model combines a subset of low-level propertieschannels, explicitly highlighting regions that are locally different.
In dynamic tension with these bottom-up influences are theinfluences of the viewer’s priorities (see Henderson, Brockmole,Castelhano, & Mack, 2007, for review). These active aspects ofvisual attention allow the viewer to select areas of the scene thatare consistent with ongoing goals. The interaction betweenbottom-up and top-down influences on visual activity are still notwell understood. However, we know that visual attention is notaffected by these influences independently1 (Adams & Mamas-sian, 2004).
One approach to examining determinants of gaze behavior is tohold the scene constant and vary the viewer’s task (Yarbus, 1967).Here we take the opposite approach: we hold the task and semanticcontent constant and vary saliency by manipulating the spatialfrequency and hence the texture2 of an image. Although a corre-lational relationship between high-spatial frequency content andthe probability of an individual directing their gaze to that regionhas been established (Mannan, Ruddock, & Wooding, 1997),causal relationship has not been adequately determined. Somestudies that have demonstrated a causal relationship (Baddeley &Tatler, 2006; Einhäuser & König, 2003) have done so in anabstract manner by utilizing global or randomly located manipu-lation in natural scenes. Such studies allow us to understand the
1 For a review of computational saliency models that attempt to incor-porate both high- and low-level influence gaze behavior; see Borji, Sihite,& Itti, 2012.
2 It has been shown that the perception of texture is heavily influencedby the spatial frequency components of an image. Similar channels withinthe saliency model have been implicated (i.e., luminance contrast: Mannan,Ruddock, & Wooding, 1997; Parkhurst, Law, & Neibur, 2002).
This article was published Online First December 23, 2013.Nida Latif, Department of Psychology, Queen’s University, Kingston,
Ontario, Canada; Arlene Gehmacher, Department of World Cultures: Ca-nadian Art, Royal Ontario Museum, Ontario, Ontario, Canada; Monica S.Castelhano and Kevin G. Munhall, Departments of Psychology and Oto-laryngology, Queen’s University.
This research was supported by funding from Natural Science andEngineering Research Council of Canada, the Brian R. Shelton Fellowship,and Queen’s University.
Correspondence concerning this article should be addressed to NidaLatif, Department of Psychology, 62 Arch Street, Humphrey Hall 307,Kingston, ON K7L 3N6, Canada. E-mail: [email protected]
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
Journal of Experimental Psychology:Human Perception and Performance
underlying mechanism of how the visual system may respond tosaliency; however, further work is needed to incorporate the in-teraction between low-level and high-level influences. More spe-cifically, expanding on findings that indicate how gaze may bedirected toward salient regions that are also semantically informa-tive (Enns & MacDonald, 2012) is important to understanding thisinteraction.
We examine the causal relationship between saliency, as indi-cated by the Itti and Koch (2001) model, and gaze by exploringnaturally occurring saliency differences in multiple copies of thesame painting (i.e., same semantic content). By maintaining se-mantic content, we demonstrate that how an observer viewedartwork was, in part, driven by saliency. Moreover, directly chang-ing saliency of semantically informative regions by manipulatingtexture causes observers to redirect their viewing of the same pieceof art.
Experiment 1
The stimulus used in this study is the historical painting TheDeath of General Wolfe (DoW), which depicts the moment ofdeath of British General James Wolfe in 1759. This paintingrepresents an event in Canadian history, a significant British vic-tory over France in Quebec during the Seven Years’ War. Featur-ing many figures both historical and allegorical, the artist Benja-min West (1738–1820) dramatized the event with such artisticvision that his work became a public sensation. High demand forthis painting resulted in West creating multiple copies, includingone commissioned by the Royal Family (von Erffa & Staley,1986). The multiple versions evidence variations, although thegeneral historical theme remained consistent. This example ofvariation in reproduction provides an effective tool to examinewhether saliency guides gaze while controlling the semantic con-tent.
Using a viewing task with constant task demands, we varied thetest image between three groups so that each group viewed adifferent version of the DoW painting. Fixations (i.e., maintenanceof gaze at a single location) were examined to investigate thedistribution of visual attention.
Methods
Participants. Forty-eight (31 women, 17 men, Mage � 20.0years) undergraduates with normal or corrected-to-normal visionparticipated for course credit or for $10/hr.
Stimuli and apparatus. Three most distinct versions fromfive available versions3 of the DoW painting were used as stimuli.Each version differs slightly in composition while preserving thesame semantic content (Figure 1B): modest alterations of thebackground, brightness and textural details and the relative loca-tion of principle figures. The three versions were selected usingtwo methods: (1) using visible differences from computed saliencymaps4 (Itti, Koch, & Niebur, 1998; Figure 1A) and (2) identifyingcompositional differences by overlaying all versions and standard-izing differences5 by aligning the central figure’s face (Supple-mental Figure S1).
The National Gallery of Canada (NG), the National Trust (NT),and the Royal Ontario Museum (ROM) versions were presented(see Figure 1) during the eye-tracking task. Five filler paintings
were also presented, including the same practice painting (Dali,1956), which was not analyzed.6
An EyeLink 1000 eyetracker (SR Research: http://www.sr-research.com/eyelink1000.html) was used, sampling at 1000Hz. All stimuli were resized to 800 � 600 resolution and weredisplayed on a 21-inch CRT monitor at a refresh rate of 100 Hz.Images subtended 38.1° � 28.6° of visual angle.
Procedure. Three groups of 16 participants each viewed adifferent version of the DoW painting along with the same fillerpaintings across groups. Participants were seated approximately 60cm away from the monitor with head motion/position controlledusing a head and chinrest.
For each trial, a central fixation cross was displayed for 500 msfollowed by an image for 10 s. Following presentation, participantsresponded to two presented questions on a scale of 1 (least) to 7(most): (1) How familiar are you with this painting? and (2) Howaesthetically pleasing do you find this painting?
Results
The gaze behavior for the three different DoW versions wasexamined to determine whether they reflected the physical differ-ences between versions. Because saliency is most predictive ofgaze in the early part of viewing an image (Underwood &Foulsham, 2006), we analyzed the first 2 s of viewing. There wasan observable correspondence between the salient regions andfixation spread and duration (Figure 1C). We observed a differencein average fixation duration among the three versions of the DoWpainting, F(2, 45) � 4.41, p � .01, �p
2 � .23. Figure 2A shows thataverage fixations were shorter for the NG version (MDiff � 55.0ms, SE � 19.7, p � .01). Because all three groups viewed the samefiller paintings, we used them as a control and found no significantdifferences in the average fixation duration (Figure 2A).
We calculated the average saccadic length to examine spread offixations. Results indicated that average saccadic length differedsignificantly between versions, F(2, 45) � 5.27, p � .01, �p
2 � .33(Figure 2B). Participants looking at the ROM version had a longeraverage saccadic length than the NG version (MDiff � 1.63°, SE �0.53, p � .01). The fillers indicated no significant differences inaverage saccadic length.
These behavioral differences did not influence participants’familiarity and aesthetic preference, and no significant differenceswere observed in participants’ subjective ratings.
Because the semantic content was maintained, Experiment 1demonstrated that the physical differences across versions were
3 The versions were as follows: National Gallery of Canada, Ottawa(1770), Royal Collection, Ickworth (1771), Clements Library, Universityof Michigan, United States (1776), National Trust, United Kingdom(1779), and Royal Ontario Museum, Toronto, Ontario, Canada (1806).
4 The Itti and Koch (2000) model creates saliency maps using a com-putational algorithm to predict where our biological system would focusattention based on a set of low-level properties including intensity, color,and orientation.
5 National Gallery of Canada (151 � 213.5 cm), Royal Collection(153.7 � 245.1 cm), Clements Library (153.7 � 244.8 cm), National Trust(152.5 � 216 cm), and Royal Ontario Museum (165 � 244.5 cm).
6 Living Still Life (Dali, 1956), The Death of Sardanapalus (Delacroix,1827), Landscape with a Long Arched Bridge (Rembrandt, 1630), Decla-ration of Independence (Trumbull, 1817), and The Wedding at Cana(Veronese, 1563).
affecting gaze patterns in a manner that can be explained bysaliency differences. Experiment 2 was conducted to directly ma-nipulate an aspect of saliency (texture) and determine if gaze canbe causally affected by this manipulation.
Experiment 2
In Experiment 2, we wanted to demonstrate that saliencyvariation was directly related to gaze differences seen in Ex-periment 1. Saliency maps demonstrated that all versions shareda common region near the main character, General Wolfe.However, the number and strength of additional regions wasgreater for the ROM version (Figure 1A). We predicted thatthese additional regions were causing participants’ gaze to bedrawn to more areas of the painting for longer periods. Weselected and manipulated the saliency maps of the ROM and theNG paintings because of their distinct saliency patterns andbehavioral responses in Experiment 1. We switched the saliency
pattern of each of the versions using sharpening and blurringfilters to simulate artistic textural modification (DiPaola, Riebe,& Enns, 2010). We predicted that switching saliency patternsthrough textural manipulation would switch gaze patterns dur-ing viewing. Our logic was as follows: our textural manipula-tion produced a change in the output of the saliency model (Itti& Koch, 2001) and thus an assumed change to prioritizedregions in observers’ visual systems. Gaze corresponding withthese prioritized regions was predicted to align with saliencydifferences.
Methods
Participants. Sixty (49 women, 11 men, Mage � 19.0 years)undergraduates with normal or corrected-to-normal vision partic-ipated for course credit or for $10/hr.
Stimuli and apparatus. The eye tracking apparatus was thesame as that used in Experiment 1.
Using the surface blurring filter and the unsharp masking7 filterin Adobe Photoshop CS5 to simulate texture,8 high-saliency re-gions were blurred and low-saliency regions were sharpened usinga masking radius of 5–30 pixels at 20%9 so the saliency map ofone painting resembled the other (see Figure 3).
A Gaussian filter with a 50-pixel radius smoothened sharpboundaries between the modified and unmodified regions to elim-inate overt indication of manipulation.10 Four versions of DoWresulted: Original NG, NG modified to have a saliency map similarto the ROM (Modified NG), Original ROM, and the ROM mod-ified to have a saliency map similar to the NG version (ModifiedROM). The same five filler paintings from Experiment 1 wereused.
Procedure. Four groups of 15 participants each viewed one offour versions of DoW along with the same filler paintings. Foreach trial, a 500-ms fixation was displayed followed by a paintingimage for 2 s. Following display of a DoW image and one otherrandomly chosen image, participants performed a free-recall taskof all details remembered about the painting.11 Participants wererecalibrated following each recall.
Results
The results from Experiment 1 were replicated in Experiment 2.There was a significant difference in average fixation duration,F(3, 56) � 3.02, p � .04, �p
2 � .12, and average saccadic length,F(3, 56) � 4.67, p � .01, �p
2 � .10, between painting versions.Individuals demonstrated longer average fixation duration(MDiff � 56.0 ms, SE � 23.1, p � .01) and greater averagesaccadic lengths (MDiff � 2.36°, SE � 0.95, p � .02) for theOriginal ROM version compared with the Original NG version. Asexpected, the switch in saliency patterns caused a switch in gazepatterns. The Modified NG demonstrated greater average fixationduration (MDiff � 53.1 ms, SE � 23.4, p � .02) and averagesaccadic length (MDiff � 3.43°, SE � 0.94, p � .01) than theOriginal NG version but showed no difference from the originalROM version that it was manipulated to resemble. The ModifiedROM showed a nonsignificant trend in fixation duration (MDiff �34.5, SE � 24.3, p � .09) compared with the Original ROMversion but showed no difference from the Original NG version itwas modified to resemble (Figure 4A and 4B).
A repeated-measures analysis of the ordinal saccadic length forthe first five fixations was performed for a more detailed exami-nation of the effect of our textural manipulation. This analysisdemonstrated a significant effect of fixation number, F(4, 45) �5.059, p � .01 (Figure 4C). By the fifth fixation, the OriginalROM and the Modified NG demonstrated a greater saccadic lengththan the Original NG and the Modified ROM, and as viewingprogressed, differences became more pronounced.
The results of Experiment 2 demonstrated that manipulatingsaliency through texture affected how participants viewed theDoW painting. Additional highly salient regions predicted theaverage saccadic length and the average fixation duration. More-over, there was a visible correspondence between gaze and therelative locations of salient regions (see Supplemental Figure S1).Reversing the saliency between the two versions caused a reversalof the gaze patterns.
General Discussion
We presented the finding that saliency and gaze are causallyrelated, a particularly striking result in an unconstrained, aestheticviewing task. Our first study demonstrated that natural variationwithin versions of the same painting produced variation in partic-
7 Surface blurring was used to preserve edges while blurring the con-tents. Hence, blurring of highly detailed regions, such as facial regions,maintained their features. Unsharp masking is used to amplify high-frequency components of an image by suppressing the low-frequencycomponents. Photographic blurring has been suggested to approximatehigh spatial frequency attenuation (Watson & Ahumada, 2011). Because ithas already been demonstrated that spatial frequency content is related totexture, we infer that these methods would manipulate texture accordingly.
8 Texture is related to a higher spatial frequency content, which can becaptured in saliency maps with the contrast and edge detection channels ofthe Itti and Koch (2000) saliency model.
9 The radius refers to the size of the edges to be enhanced or dampened.The range used accommodates both small- and large-scale manipulations.The amount listed as a percent controls the magnitude of contrast appliedto the edges.
10 Gaussian blurring is not detected by the Itti and Koch (2000) saliencymodel and thus did not influence the saliency measures.
11 The results for this memory task will not be presented in this article.
Figure 2. (A) Average fixation duration. Fixation duration for two groupsviewing the National Trust (NT) and Royal Ontario Museum (ROM)version was significantly longer than the National Gallery of Canada (NG)version. There were no differences in fixation durations for the average ofthe filler paintings among the three groups. Error bars � SE. (B) Averagesaccadic length. The NT and ROM versions indicated significantly longersaccadic length than the NG version. No differences were seen in theaverage saccadic length for the fillers. Error bars � SE.
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
36 LATIF, GEHMACHER, CASTELHANO, AND MUNHALL
ipants’ gaze patterns. Our second study showed that manipulationof an aspect of saliency caused participants’ gaze to change sys-tematically.
Experiment 1 showed that gaze variations caused by physicaldifferences in the paintings were consistent with the visual sa-liency hypothesis: our visual system is sensitive to differences inlow-level dimensions such as color, intensity, and edge orientation(Itti & Koch, 2000), and these properties are predictive of saccadicmovements. The behavior observed in Experiment 1 cannot beassociated with the cognitive scene processing (e.g., Henderson etal., 2007) because the semantic content differences are subtle; alow-level, property-driven explanation is more appropriate. Wefurther supported this hypothesis by using the Itti and Koch (2000)saliency model to demonstrate that the versions were different insaliency content and suggested that gaze differences were a resultof the variation.
In Experiment 2, we used textural manipulation to explorewhether image saliency directly affected how participants viewedthe paintings. Textural manipulation was selected because high-texture areas (where the spatial frequency is relatively greater thansurrounding regions) have shown to direct gaze in natural scenes(Baddeley & Tatler, 2006). In addition, many artists have indicatedtheir awareness of using texture to guide viewers’ gaze toward thecenter of focus, an effective tool because highly textured areas areoften associated with more meaningful aspects of a scene (Hen-derson et al., 2007). We demonstrated this association by showingthat naturally occurring salient regions, and hence, our manipula-tions were not random; they corresponded with semantically in-formative regions such as various characters that may be moremeaningful. This is consistent with the finding that high spatialfrequency regions are prioritized when perceiving semanticallymeaningful information (Enns & MacDonald, 2012).
Figure 3. Original and modified saliency maps for the National Gallery of Canada (NG) and the Royal OntarioMuseum (ROM) versions along with corresponding gaze patterns. (A) Gaze patterns and saliency patterns forthe original NG and ROM versions of the painting. Size of circles indicate duration of fixations. (B) Gazepatterns and saliency maps for the version of NG modified to look like the ROM and the ROM modified to looklike the NG version. The size of circles indicate duration of fixation. Note the switch in patterns with texturalmodification. The data displayed for all four groups were for the first five fixations of the experiment.
Thi
sdo
cum
ent
isco
pyri
ghte
dby
the
Am
eric
anPs
ycho
logi
cal
Ass
ocia
tion
oron
eof
itsal
lied
publ
ishe
rs.
Thi
sar
ticle
isin
tend
edso
lely
for
the
pers
onal
use
ofth
ein
divi
dual
user
and
isno
tto
bedi
ssem
inat
edbr
oadl
y.
37ART OF GAZE GUIDANCE
The textural component of an image is also involved in therecognition of stimuli; texture and clarity may direct attention tofamiliar objects. This confluence of high texture and importancemay result in an aesthetic preference and attention to these regions(Zeki, 1999). In terms of art perception, we may prioritize suchregions because they signal where an observer ought to look from
the perspective of the artist (Enns & MacDonald, 2012). Accord-ingly, results of Experiment 2 confirmed the causal relationshipbetween gaze and texture. By switching the saliency patternsbetween versions, we were able to switch the gaze behavior.
As a caveat, although we demonstrate causality, at no point dowe suggest that bottom-up influences are completely independent.Instead, we focused our investigation based on our current knowl-edge of high-level influences in scene perception, that is, by usingthe highly cognitive task of aesthetic evaluation (Zeki, 1999) andusing semantically coupled saliency manipulations (Enns & Mac-Donald, 2012). Further investigation of the interaction betweenthese influences is required. Returning to the results, these data begthe question of why the paintings vary in saliency. Artists are ableto present the visual world in a manner that exploits the humanvisual system. However, our results do not allow us to confirmWest’s intentionality in introducing differences in these particularpaintings. In addition to artistic composition, there may be otherexplanations for appearance and saliency alteration: aging underdifferent conditions and multiple instances of restoration/preser-vation by both the artist and subsequent caretakers (e.g., removal/addition of varnish, cleaning and paint-loss filling; Daly-Hartin &Vogel, 2003). Although the source of differences is unclear, thedifferences in gaze patterns caused by textural modification areapparent.12 Further historical research is needed to understand theartist’s true intent, but based on our data, we suggest that artistscan use artistic techniques to guide art viewing much like ourexperimental manipulation.
Our studies demonstrated how an artist could exploit the oper-ating principles of the visual system by contrasting artistic tech-niques used to modify saliency with the high-level features of ascene’s meaning. Artistic compositions provide an opportunity tostudy scene perception in a manner naturally unbiased by explicitpredetermined tasks, though not excluding implicit goals of aes-thetic appreciation and scrutiny. Even in the relatively high-levelcognitive task of art observation, visual attention is influenced bythe low-level properties of saliency. Our demonstrated preferencefor such salient features may intuitively compel artists to usetechniques to direct the gaze of their audience.
12 We note the increased saliency on the character Brigadier RobertMonckton in the ROM version, modified by West in 1806, and the possiblecoincidence with the eventual acquisition of the painting by the Monktonfamily in 1829 (von Erffa & Staley, 1986). Brigadier Robert Monckton isfeatured in the red army uniform in the center of the group located to theleft.
References
Adams, W. J., & Mamassian, P. (2004). The effects of task and saliency onlatencies for colour and motion processing. Proceedings of the RoyalSociety: B: Biological Sciences, 271, 139–146. doi:10.1098/rspb.2003.2566
Baddeley, R. J., & Tatler, R. W. (2006). High frequency edges (but notcontrast) predict where we fixate: A Bayesian system identificationanalysis. Vision Research, 46, 2824–2833.
Borji, A., Sihite, D. N., & Itti, L. (2012). Quantitative analysis of human-model agreement in visual saliency modeling: A comparative study.IEEE Transactions on Image Processing, 55–69.
Daly-Hartin, D., & Vogel, P. (2003, June 3). The rewarding challenges oftreating the Death of General Wolfe. Retrieved from the Canadian
Figure 4. (A) Average fixation duration for the two original and twomodified versions of the The Death of General Wolfe (DoW) paintings.Error bars � SE. (B) Average saccadic length for the two original and twomodified versions of the DoW paintings. Error bars � SE. (C) Averageordinal saccadic length for the first five fixations in the original andmodified versions of the DoW paintings. Error bars � SE.
Conservation Institute website: http://www.cci-icc.gc.ca/caringfor-prendresoindes/articles/361-eng.aspx
DiPaola, S., Riebe, C., & Enns, J. T. (2010). Rembrandt’s textural agency:A shared perspective in visual art and science. Leonardo, 43, 145–151.doi:10.1162/leon.2010.43.2.145
Einhäuser, W., & König, P. (2003). Does luminance-contrast contribute toa saliency map for overt visual attention? European Journal of Neuro-science, 17, 1089–1097. doi:10.1046/j.1460-9568.2003.02508.x
Enns, J. T., & MacDonald, S. C. (2013). The role of clarity and blur inguiding visual attention in photographs. Journal of Experimental Psy-chology: Human Perception and Performance, 39, 568–578.
Henderson, J. M., Brockmole, J. R., Castelhano, M. S., & Mack, M. (2007).Visual saliency does not account for eye movements during visualsearch in real-world scenes. In R. van Gompel, M. Fischer, W. Murray,& R. Hill (Eds.), Eye movements: A window on mind and brain (pp.538–562). New York, NY: Elsevier.
Itti, L., & Koch, C. (2000). A saliency-based search mechanism for overtand covert shifts of visual attention. Vision Research, 40, 1489–1506.doi:10.1016/S0042-6989(99)00163-7
Itti, L., & Koch, C. (2001). Computational modeling of visual attention.Nature Reviews Neuroscience, 2, 194–203. doi:10.1038/35058500
Itti, L., Koch, C., & Niebur, E. (1998). A model of saliency-based visualattention for rapid scene analysis. IEEE Transactions on Pattern Anal-ysis and Machine Intelligence, 20(11), 1254–1259.
Mannan, S. K., Ruddock, K. H., & Wooding, D. S. (1997). Fixationpatterns made during brief examination of two-dimensional images.Perception, 26, 1059–1072.
Parkhurst, D., Law, K., & Neibur, E. (2002). Modeling the role of saliencein the allocation of overt visual attention. Vision Research, 42(1), 107–123.
Underwood, G., & Foulsham, T. (2006). Visual saliency and semanticincongruency influence eye movements when inspecting pictures. TheQuarterly Journal of Experimental Psychology, 59, 1931–1949.
von Erffa, H., & Staley, A. (1986). The paintings of Benjamin West. NewHaven, CT: Yale University Press.
Watson, A. B., & Ahumada, A. J. (2011). Blur clarified: A review andsynthesis of blur discrimination. Journal of Vision, 11. doi:10.1167/11.5.10
Yarbus, A. L. (1967). Eye movements and vision. New York, NY: PlenumPress. doi:10.1007/978-1-4899-5379-7
Zeki, S. (1999). Art and the brain. Journal of Consciousness Studies, 6,76–96.
Received May 8, 2013Revision received September 11, 2013
