Understanding the Historic Emergence of Diversity in Painting via Color Contrast

4.4 MB·

by  Byunghwee Lee

UnderstandingtheHistoricEmergenceofDiversityinPaintingviaColorContrast
Byunghwee Lee,1 Daniel Kim,2 Hawoong Jeong,1,3,4 Seunghye Sun,5 and Juyong Park6,7,∗
1DepartmentofPhysics,KoreaAdvancedInstituteofScienceandTechnology, Daejeon34141,Korea
2NaturalScienceResearchInstitute,KoreaAdvancedInstituteofScienceandTechnology,Daejeon34141,Korea
3InstitutefortheBioCentury,KoreaAdvancedInstituteofScienceandTechnology,Daejeon34141,Korea
4APCTP, Pohang, Gyeongbuk 37673, Korea
5Directorate of Culture and Arts Research, The Asia Institute, Seoul 04310, Korea
6GraduateSchoolofCultureTechnology,KoreaAdvancedInstituteofScienceandTechnology,Daejeon34141,Korea
7BK21 Plus Postgraduate Program for Content Science, Daejeon 34141, Korea
Paintingisanartformthathaslongfunctionedasmajorchannelforcommunicationandcreativeexpression.
Understandinghowpaintinghasevolvedoverthecenturiesisthereforeanessentialcomponentforunderstand-
ingculturalhistory,intricatelylinkedwithdevelopmentsinaesthetics,science,andtechnology. Theexplosive
growthintherangesofstylisticdiversityinpaintingstartinginthenineteenthcentury,forexample,isunder-
7 stood to be the hallmark of a stark departure from traditional norms on multidisciplinary fronts. Yet, there
1 existfewquantitativeframeworksthatallowustocharacterizesuchdevelopmentsonanextensivescale,which
0 wouldrequirebothrobuststatisticalmethodsforquantifyingthecomplexitiesofartisticstylesanddataofsuf-
2 ficientqualityandquantitytowhichwecanfruitfullyapplythem. Hereweproposeananalyticalframework
thatallowsustocapturethestylisticevolutionofpaintingsbasedonthecolorcontrastrelationshipsthatalso
n
a incorporatesthegeometricseparationbetweenpixelsofimagesinalarge-scalearchiveof179853images. We
J firstmeasurehowpaintingshaveevolvedovertime,andthencharacterizetheremarkableexplosivegrowthin
diversityandindividualityinthemodernera.Ouranalysisdemonstrateshowrobustscientificmethodsmarried
5
with large-scale, high-quality data can reveal interesting patterns that lie behind the complexities of art and
2
culture.

]
V
INTRODUCTION artifactsthatprovidesanunprecedentedopportunitytodevise
C
science-inspiredanalyticalmethodsforidentifyinginteresting
.
s Humans have used paintings as a way to communicate, andcomplexpatternsresidinginartandcultureenmasse[2–
c
[ recordevents,andexpressideasandemotionssincelongbe- 7].
foretheinventionofwriting. Paintinghasbeenatthecenter The quantitative study of style in a cultural artifact is also
1
of artistic and cultural evolution of humanity reflecting their called stylometry, a term coined by Polish linguist Wincenty
v
4 lifestyle and beliefs. For example, Ernst Gombrich, art his- Lutosławski who attempted to extract statistical features of
6 torian in Britain, remarked on Egyptian art in his book that word usage from Plato’s Dialog [8]. Stylometric analyses
1 “To us these reliefs and wall-paintings provide an extraordi- havebeenperformedonvarioussubjectssincethen,including
7 narily vivid picture of life as it was lived in Egypt thousand literature[9–12],music[13–16]andart[17–24]. Alandmark
0 years ago” [1]. A deep investigation into how painting has scientificstudyofpaintingsisTayloretal.’scharacterization
1. evolved and the motivations behind it, therefore, can be ex- ofthefractalpatternsinJacksonPollock’s(1912–1956)drip
0 pected to yield valuable insights into the history of creative paintings [17]. It was subsequently found that the character-
7 developmentsinculture. istics of drip paintings of unknown origin significantly devi-
1 Giventheubiquityofartandculture,andthevalueourso- atedfromthoseofPollock’s,showingthatsuchmeasurements
v: cietyputsonthemassymbolsofthequalityoflife,webelieve can reflect an artist’s unique style [18]. Other notable stud-
i thatapproachingartandcultureassubjectsofseriousscience ies of paintings include Lyu et al.’s which decomposed im-
X
shouldbeaworthyendeavor. Inordertoproceed,wetakethe agesusingwavelets[19]; Hughesetal.’swhichusedsparse-
ar viewpointthata pieceofartcanbeconsideredasacomplex coding models for authenticating artworks [20]; and Kim et
systemcomposedofdiverseelementswhosecollectiveeffect, al.’s which studied the “roughness exponent” to characterize
whenpresentedtoanaudience,istostimulatetheirsenses— brightnesscontrasts[21]. Venturingbeyondquantificationof
be they cerebral, emotional, or physiological. To understand artistic styles, recent studies investigated perceived similari-
a painting, for instance, one may analyze its colors, geome- ties between different paintings [22], the influence relation-
try,brushingtechniques,subjects,orimpactontheaudience, ships between artworks for quantifying creativity in an art-
each of which would allow us to grasp the multifacted, cor- work[23],andthechangesintheperceptionofbeautyusing
related aspects of the art form. The same could be said of face-recognitiontechniquesonimagesfromdifferentera[24].
manyotherartforms,obviouslywithsomeuniquevariations
depending on the form. A positive development that could Despite these attempts, individual stylistic characteristics
have far-reaching benefits for such work is the recent pro- ofpaintershavenotyetbeensufficientlyandcollectivelyex-
liferation of high-quality, large-scale digital data of cultural plored, which will also reveal a remarkable diversity in the
modern times. This stem from a number of shortcomings of
previous works: They lack robust statistical frameworks for
understanding the underlying principles based on the image
∗Correspondingauthor;[email protected] data;theymakeonlylimiteduseofthefullcolorinformation,
2
(a) (b)
&)
g (
o
l
pixels
pixels
(c) (d)
&)
g (
o
l
pixels
pixels
FIG.1. Thecolorcontrastofapaintingisquantifiedusingtheinter-pixelcolordistances. Theinter-pixelcolordistancesarevisualizedalong
thez-axis,overlaidontheimages.(a)PietMondrian’sCompositionA.Inthepainting,dissmallonaveragebutasignificantnumberoflarged
existleadingtoahighlyheterogeneousdistributionofd.Theyaremanifestedastheconspicuous“walls”alongtheboundariesbetweenareas
ofrelativelyuniformcolors. (b)DistributionofcolordistancesdintheMondrianinthelog-linearscales. Theblackdots‘+’representthe
randomizationgeneratedbyshufflingthelocationsofthepixels,averagedover100samples(Randomized1). Thegraydots‘×’representthe
randomizationgeneratedbyreplacingallpixelsintheoriginalimagewithuniformlysampledfromcolorsfromRGBcolorspace(Randomized
2). (c)ClaudeMonet’sWaterLiliesandJapaneseBridge. Notethelackofwell-separatedwallsthatseparatecolorpatchesasseeninthe
Mondrian. (d)TheMonetshowsahighaveraged owingtothemanyintertwiningbrushstrokesofdifferentcolors,butfewextraordinarily
larged,resultinginarapidlydecayingtailofthedistributionofd,similartoanexponential.
even though it is readily available in data; or they concern colorsarejuxtaposedtoforma‘hardedge’painting,apopular
themselves with specific artworks or painters. In this work, styleofthetwentiethcentury.
we propose a scientific framework for characterizing artistic Weproposeaquantitywecallseamlessnessforcolorcon-
styles that makes use of the complete color profile informa- trastthatincorporatesthecolorprofileandgeometriccompo-
tion in a painting and simultaneously takes into account the sitionofapainting. Weshowthatthisquantityisausefulin-
geometrical relationships between the colored pixels in the dicator for characterizing distinct painting styles, which also
image, two essential building blocks of an image. Applying allows us to track the stylistic evolution of western painting
this framework to a large number of historical paintings, we both on the aggregate and the individual levels (particularly
characterizeartisticstylesofvariouspaintingsovertime. formodernpainters)byapplyingittoatotalof179853dig-
italscansofpaintings—thelargestyetinourtypeofstudy—
Colorboastsalonghistoryasasubjectofintensivescien-
collectedfrommultiplemajoronlinearchives.
tific investigation from many points of view including phys-
ical, physiological, sensory, and so forth. Starting with two
classical groundbreaking investigations by Newton [25] and
Goethe [26, 27], modern research on color continues in full DATADESCRIPTION
forceinart,biology,medicine,andvisionscience[28,29]as
wellasphysics. Hereweintroducetheconceptof‘colorcon- We collected digital scans of paintings (mostly western)
trast’asthesignaturepropertyofcoloruseinapainting.Asits from the following three major online art databases: Web
namesuggests,colorcontrastreferstothecontrasteffectorig- GalleryofArt(WGA)[3],WikiArt(WA)[4],andBBC-Your
inating from the color differences between clusters of colors Paintings (BYP) [5]. The WGA dataset contains paintings
inapainting.Examplesofpaintingsinwhichcolorcontrastis dated before 1900, while the WA and BYP datasets contain
highlypronouncedincludeVincentvanGogh’s(1853–1890) those dated up to 2014 (all datasets are up-to-date as of Oct
StarryNight(1899)whereabrightyellowmoonfloatsagainst 2015). WGA provides two useful metadata on the paintings
thedarkblueskyandPietMondrian’s(1872–1944)Composi- allowingadeeperanalysis: paintingtechnique(e.g.,tempera,
tionA(1923)wherewell-definedgeometricshapesofdistinct fresco,oroil)andpaintinggenre(e.g.,portrait,stilllife,genre
3
(a) (b) (c)
1500 K Original 10000 K
FIG. 2. Images (Jean-Franois Millet, The Gleaners (1857)) under various simulated temperatures. (a) 1500 K (close to candle light), (b)
originalimage,(c)10,000K(closetoclearbluesky).
(a) (b)
P1 P2 P3 P4 P5 P6
FIG. 3. S of six sample images under different (a) color temperatures, and (b) image sizes. The sample images are P1: Piet Mondrian,
CompositionA(1923),P2: Caravaggio,St. FrancisinEcstasy(c.1595),P3: WillemKalf,StillLifewithDrinking-Horn(c.1653),P4: Pieter
Bruegel the Elder, The Census at Bethlehem (1566), P5: Jean-Franois Millet, The Gleaners (1857), P6: Claude Monet, Water Lilies and
JapaneseBridge(1899).
painting—itself a specific genre depicting scenes from ordi- METHODOLOGY
nary life). BYP is mainly a collection of oil paintings pre-
servedintheUnitedKingdomwheremostpaintingsoriginate
ColorcontrastandseamlessnessS asitsmeasure
from. (LaterweshowthatBYPdatastillexhibitacompara-
bletrendincolorcontrastwithotherdatasets.) Weexcluded
As its name suggests, color contrast represents the effect
paintings dated before the 1300s, as they were too few. We
broughtonbycolordifferencesbetweenpixelsinapainting.
manuallyremovedthosedeemedimproperfor,oroutsidethe
Therefore, characterizing how an artist places various colors
scopeof,ouranalysis;theyincludepartialimagesofalarger
on a canvas is a key element in determining the color con-
original, non-rectangular frames, seriously damaged images,
trastinapainting.Thehumansenseofcolorcontrastbetween
photographs,etc.Thefinalnumbersofpaintingusedforanal-
twopointsinapainting(pixelsinadigitalimage)isaffected
ysis are 18321 (WGA), 70235 (WA), 91297 (BYP) images
moststronglybytwofactors,therelativecolordifferencebe-
foratotalof179853.
tween the pixels and their geometric separation—the closer
theyareinrealspacethemorepronouncedtheircolordiffer-
encewillbe. Inordertoquantifythisphenomenon,wemust
firstdefinethecolordifferencebetweentwopixelsthatagrees
withthehuman-perceiveddifference. Thenwemustincorpo-
4
1.0 tributions in d indicates that artists are likely to use similar
WGA colors, avoiding extremely distant colors in the color space
WA (Fig. 1(b), (d)). Furthermore, when geometry is considered,
0.8
BYP thesimilarcolorsalreadyinasamepaintingtendtostayclose
in real space also. Therefore, the final distribution of d of
F0.6 a painting can be considered as the signature of a painting
D
that represents both an artist’s own color selection and geo-
C
C metricstyle. Animagecharacterizedbyahighcolorcontrast
0.4
showsregionswithlargeinter-pixelcolordistancerelativeto
theoverallimage,i.e. inhomogeneityind. IntheMondrian,
0.2
d is small on average but a significant number of large d ex-
ist, namely the tail of π(d) decays more slowly than an ex-
0.0 ponential (Fig. 1(b)). This is a consequence of large patches
0 500 1000 1500 2000 2500 3000 3500 4000
ofnearlyuniformcolorsbeingseparatedbywell-definedbor-
Width (pixels)
ders. The Monet, on the other hand, shows a high average
d owing to the many intertwining brush strokes of different
FIG.4.Thecomplementarycumulativedistributionofimagewidths colors,butfewextraordinarilylarged,similartoanexponen-
(thelongersideofanimage).
tial distribution (Fig. 1(d)). These suggest using the relative
magnitude of the mean d¯ and the standard deviation σ to
d
characterize a painting’s overall color contrast: Specifically,
ratethespatialseparationinformationtoproduceacombined weuseS ≡ (σ −d¯)/(σ +d¯). S ∈ [−1,1]andS > 0when
d d
measure. σ >d¯(S =0.46fortheMondrian),andwhen(S =−0.12for
d
Acolorisrepresentedbythreevaluesthatdefinethethree the Monet). We call it “seamlessness” because a high (low)
coordinatesina‘colorspace’.Acolorspaceisnamedaccord- S means fewer (more) boundaries or ‘seams’ between clus-
ingtowhatthecoordinatesmean. Familiarexamplesinclude tersor‘patches’oflikecolors. Thisquantityisalsousedfor
theRGBspaceforRed,Green,andBlue,theHSVspacefor quantifying heterogeneity of inter-event time distributions in
Hue(thecolorwheel),Saturation,andValue(brightness),and statisticalphysics,althoughtheproblemsareunrelatedtoeach
theCIELabspace(thefullnomenclatureis1976CIEL∗a∗b∗) other[31].
forL∗(lightnessrangingfrom0forblackto100forwhite),a∗
(running the gamut between cyan and magenta), and b∗ (be-
tweenblueandyellow). Thea∗ andb∗ axeshavenospecified
RobustnessofS
numericallimits. ForourworkweusetheCIELabspaceasit
wasdesignedspecificallytoemulatethehumanperceptionof
differencebetweentwocolorswhichisproportionaltotheEu- Unlike digitized or OCR’ed (Optical Character Recogni-
(cid:113) tion) text, a digitized image of a painting can exist in many
clidean distance, d = (L1∗−L2∗)2+(a∗1−a∗2)2+(b∗1−b∗2)2 versions of differing sizes or colors depending on scanning
betweenthe(L∗,a∗,b∗)coordinatesoftwocolors[30]. environments and settings. We therefore need to test for the
Thatacolordifferencedbetweenpixelswouldbethemore robustness (insensitivity) of S against such variations, if we
pronounced the closer they are prompts us to consider, for aretobeabletorelyonitasacharacteristicofapainting,and
simplicity, that between adjacent pixel pairs. This results in notonlyofaspecificscanofit. Whileweexpectslightdiffer-
atotalofapproximately2N datapointstoconsiderinanim- encesincolororresolutionnottoresultinsignificantchanges
age of N pixels. To illustrate what the d can teach us about inS inprinciplesinceitisdefinedintermsofthecolordiffer-
the use of color in a painting, we compare the distribution ences between pixels, it would still be reassuring to confirm
π(d)forPietMondrian’s(1872–1944)CompositionA(1923) therobustnessofS againstcertainrealizablevariations.
(Fig.1(a))andClaudeMonet’s(1840–1926)WaterLiliesand Whenonedigitallyscansapainting,thelightingcondition
Japanese Bridge (1899) (Fig. 1(c)), which show significant isakeyelementthataffectsthefinalcoloroftheimage. Since
differencesthatindeedwellreflecttheirvisibledifferences. In the original lighting conditions are not given in the datasets,
order to provide a baseline for proper comparison, we also wesimulatedifferentlightingconditionsbyvaryingthecolor
measurethecolordistancedistributionoftwodifferenttypes temperatures of the light sources, i.e. the color profile of a
of null models obtained from randomizing these paintings. blackbodyofthesametemperature[32]. Assumingthatthe
The first null model is produced by randomly relocating the original scan represents a well white-balanced image, multi-
pixels of the original image while preserving the number of plyingeachpixelbytheRGBvaluesofthecoloroftheblack
each color, and the second is produced by replacing all pix- body normalized by 255 (the maximum value of each axis)
els of the original image with randomly selected colors in gives the simulated pixel. For instance, at 1500 K, the RGB
the RGB color space. Therefore, the first randomization re- value of the color that a black body radiates is (255, 109,
tains the intrinsic color of a painting with only its geomet- 0). Then the pixels of an image are multiplied by the fac-
ricstructuredestroyed,wherethesecondrandomizationpro- tor(255/255,109/255,0/255). Analysisofthesixtestimages
duces a completely random image. The fact that the second in different conditions of light ranging from 1500 K (similar
null model shows a significantly broader tail than other dis- toacommoncandle)to10000K(similartoaveryclearblue
5
FIG.5. EvolutionofS overtime. (a)Numberofpaintingsinthethreedatasetsused. (b)S ofallpaintingsinthedatasetsfrom1300CEto
2014CE.AsuddenbroadeninginS isobservedstartinginthemid-nineteenthcentury.(c)EvolutionofaverageS overtime,withthestandard
errorofthemeanindicated.(d)GrowthofstandarddeviationofS.Thestandarddeviationgraduallyrisesthroughouthistory,withthelargest
leapobservedbetweenthenineteenthandthetwentiethcenturies. (e)Increaseinpainters’diversityrepresentedbythestandarddeviationof
individuals’S values. Eachgraydotindicatesanartist. (f)Thenumberofpaintingsproducedbyanartistshowsnocorrelationwithdiversity
(PearsonCorrelationCoefficient=0.02).
sky)indicatesthatS isfairlyconsistentasexpected(seeFig.2 the six test images to between 100 and 1500 pixels in width
andFig.3(a)forthesimulatedimagesofpaintingsindifferent (the longer side of an image) using the bicubic interpolation
colortemperatureandtheirS values). method(Fig.3(b)). Aftershowingsomefluctuationwhenthe
sizeoftheimageisverysmall(<300pixelsinwidth),S be-
WealsotestfortheeffectofimagesizeonS byrescaling
6
(a) (b) modeling[33])duringtheRenaissanceperiod,andtenebrism
(painting in the shadowy manner with dramatic contrasts of
lightanddark[33]byCaravaggio(1517–1610)madepopular
duringtheBaroqueperiodcontributedtotheincreaseinS.
The development of these new painting techniques is also
closely related to the rise of novel painting genres. The rise
in popularity of portraits after the fifteenth century led to a
(c) (d) coupleofsignificantdevelopmentsinpaintingtechniquesuch
as chiaroscuro and tenebrism mentioned above (Fig. 6(c)).
Stilllifeshowsnotablechangesduringthesixteenthcentury,
reachingthepeakintheseventeenthcentury(Fig.6(d)). The
increaseofS instilllifeinthesixteenthcenturyisattributed
tothechangeofthemesandsubjects. Inthepriorhalfofthe
century, Dutch painters like Pieter Aertsen (1508–1575) and
JoachimBeuckelaer(1533–1573)intentionallycombinedstill
lifeanddepictionofbiblicalscenesinthebackground,while
in the latter half artists began to highlight still objects by in-
corporatingchiaroscuropreviouslyfoundinportraitsresulting
FIG.6. EvolutionofSofdifferentpaintingtechniquesandgenres.
highS [33].
(a)NumbersofpaintingsofvarioustechniquesintheWGAdataset.
(b)EvolutionofSofdifferentpaintingtechniques,withthestandard
errorofthemeanindicated. KS-testsupontheshadedareaconfirm
Inthenineteenthcentury,artistsbegantoperceivepaintings
thatthedistributionofSofdifferenttechniquesaresignificantlydif-
as a means of expressing their individuality and originality
ferent(P < 10−11 forallpairs). (c)Numberofpaintingsinvarious
more strongly than ever before [34]. Challenging the tradi-
genresintheWGAdataset. (d)EvolutionofSofdifferentgenres,
tionledtoathrivingofdifferentinterpretationsoftheworld,
withthestandarderrorofthemeanindicated.
andvariousnewtechniquesforexpressingitemerged[1]. In
the beginning of the nineteenth century, for instance, artists
comes fairly stable for larger widths (> 500 pixels). Since started to pursue various impressions of light shining on na-
99.8% (WGA), 76.0% (WA) and 100.0% (BYP) of painting ture and landscapes, rather than the dramatic and artificial
images in this research are wider than 500 pixels, this is un- lighting effect of the previous era, leading to the decrease
likelytobeanissueinpractice(Fig.4). in S. The invention of the railroad and the paint tube en-
abled impressionists to travel to distant areas, leading to the
surge in popularity of landscape paintings in the nineteenth
RESULTS century [35] (Fig. 6(c)). Furthermore, towards the end of
the nineteenth century modern abstract art began to emerge,
noted for a drastic departure from realism [1]. The simple
Historicalevolutionofcolorcontrast
andgeometricabstractionofthemovementledtoanunprece-
dented growth in S (Fig. 5(c)). But it is important to note
The measurement of S on all images allows us to map
thatthevarianceinS alsogrowsrapidly,indicatingaremark-
the historical trend of color contrast, shown in Fig. 5(b).
able growth in the diversity in color contrast. The most no-
Most notably, the average S consistently increases until it
tablegrowthoccursbetweenthenineteenthandthetwentieth
shows a temporary dip in the nineteenth century (Fig. 5(c)).
centuries(Fig.5(d)). Fig.7showsthatinearlierperiods,the
The increase in S around the fifteenth century is often at-
shape of distribution of S is concentrated around the mean,
tributed to the wide adoption of oil as binder medium for
reflecting a narrow scope of color usage and therefore color
pigments [1, 33]. The availability of new pigments, me-
contrast. However, in modern periods, the distribution be-
dia, and colors have historically been linked to the emer-
comesmuchbroaderthanearlierperiods. Thisindicatesthat
gence of new techniques and styles in painting. Prior to
paintingstylesbecomemorediverseinlatertimes,especially
1500 CE, in the Medieval times, most paintings were tem-
the modern era. The concentration around the mean appears
pera or fresco. Around the fifteenth century, oil gained pop-
weak,makingitdifficulttothinkofa“typical”style.
ularity, superseding the previous two as the most dominant
medium of choice that allowed for new techniques for high
contrast (Fig. 6(a)). Fig. 6(b) teaches us that oil paintings Thisstrongerdiversityincolorcontrastisobservednotonly
show significantly higher average S than other techniques. on such aggregate level, but also in individual painters’ pro-
TheKolmogorov-Smirnovtesttellsusthatthedistributionsof files: Regardless of the numbers of paintings produced, the
S ofvaryingtechniquesaresignificantlydifferent(P < 10−11 individual painter exhibits a wider range of S in this period
forallpairs). Additionally, twowell-knownhistoricaldevel- thantheirpredecessors(Fig.5(e),(f)),signifyingacultureof
opments in painting, the chiaroscuro (the treatment of light experimentationandwillingadoptionofdiversestyles[1].We
anddarktoexpressgradationsoflightthatcreatetheeffectof explorethisinmoredetailinthefollowingsection.
7
(a) (b) (c)
FIG.7.ThechangingvariancesofS overtime((a))WA,(b)WGA,(c)BYP).Thedistributionsarethebroadestinthemodernera.TheWGA
datasetcontainspaintingsonlyupto1900.
Characterizingindividualpaintersinthemodernera notable, prominent ones. Fig. 9(a) shows 100 modern artists
whose µ is ranked in the top 50, both positive and nega-
Prompted by the aforementioned extraordinary historical tive. It is American painter Howard Mehring (1931–1978)
developments in color contrast in the modern, we find it es- who has the largest µ = 4.07. Mehring’s early works are
sential to explore in finer detail the patterns of individual- reminiscent of Pollock, Mark Rothko (1903–1970) and He-
ityforunderstandingstylisticdevelopments. Forthemodern len Frankenthaler (1928–2011), employing uniformly scat-
painters who belong to this period (defined as those whose teredcolorswithvagueboundaries[36]. Hislaterworks, on
middle point in their career is in the nineteenth century or theotherhand,becomemorestructuredwithgeometriccom-
later),weintroducetwonovelquantitiestocharacterizetheir positions of vivid colors with abrupt transitions, very simi-
individuality, metamorphosality and singularity. Metamor- lar to Mondrian’s hard-edge paintings. At the other extreme
phosalitymeasuresapainter’stransformationincolorcontrast with the smallest (most negative) µ is Swiss-French painter
overtheircareer,whilesingularitymeasureshowdistincttheir Félix Edouard Vallotton (1865–1925), member of the post-
styleisfromthenormoftheday. WeusedtheWAdatasetto impressionist avant-garde group Les Nabis; initially famous
measurethesequantities. forwoodcutsfeaturingextremelyreductiveflatpatternswith
strong outlines (high S), he produced classical-style paint-
ings such as landscapes and still life in later life (low S) for
µ=−5.59.
Metamorphosality
Mondrian,founderofDeStijlmovementandrenownedfor
abstractionist paintings (Fig. 8(d)), actually produced works Singularity
of a wide range of styles over his career. His progression
from traditional style to abstractionism can in fact be sum- Another indication of a strong individuality is how one’s
marized using S, which increases consistently until the mid- works differ from their contemporaries’. We quantify this
1920s,whenhisabstractionismfullymatures(Fig.8(a),(d)). using singularity defined as follows. For each painting we
Pierre Auguste Renoir (1841–1919), an early leader of im- compare its S with those produced roughly at the same time
pressionism, is the opposite: his S decreases over time, as (defined as a span of eleven years, five years before and five
heprogressivelyemploysfree-flowingbrushstrokestogener- afteritsdate)andmeasureitsz-score. Wecallapaintingsin-
ateboundariesthatfusesoftlywiththebackground(Fig.8(b), gular (i.e. statistically unusual) if its S falls outside some
(e)). Claude Monet (1840–1926) and Edgar Degas (1834– |z| value, which we take to be |z| > 1. We then measure
1917), also prominent impressionists, show similar trends. eachpainter’sproductionrateofsingularartworksovertheir
These observations teach us that the changes in S can in- careers. Fig. 8(f) shows seven artists and their paintings’ z-
deed represent painters’ stylistic evolutions. We now define scores,forexample. Theartworksinthelightlyshadedareas
themetamorphosalityofapainterbasedontheslopeaofthe arethesingularonesaccordingtoourdefinition. Wenowde-
linearfittotheS valueswiththeircareerlengthsnormalized finethesingularityνofanartistasthedifferencebetweenthe
to 1. For instance, a = 0.62 for Mondrian and a = −0.10 fractions of their paintings that are z > 1 and z < −1. This
for Renoir (Fig. 8(c)). Given the near-Gaussian distribution definition allows us to determine those who often produced
π(a) over the 1,326 modern artists who produced paintings singular paintings, and show a specific trend in S. For ex-
in at least five distinct years, we define a painter’s metamor- ample, 45% of Mondrian’s paintings are in z > 1 (singular
phosality as their z-score µ ≡ (a−a¯)/σ , where a¯ is the av- high in S) and only 6% in z < −1, resulting in ν = 0.39,
a
erage, and σ is the standard deviation of the slopes. This showingthathishigh-Spaintingsareindeeduniqueandsin-
a
allows us, for example, to rank the artists and find the most gular when compared with his contemporaries. In Fig. 8(g)
8
(a) (b) (c)
a = +0.62 M6 a = -0.10
M5
M7 R2
R1 R4
M3
M4
M2 R5 R6
R3 R7
M1
(1895) (1944) (1860) (1919)
(d) (e)
Piet Mondrian Pierre Auguste Renoir
M1 (1907) M2 (1911) M3 (1912) R1 (1870) R2 (1872) R3 (1876)
M4 (1918) M5 (1923) M6 (1936) M7 (1943) R4 (1880) R5 (1890) R6 (1902) R7 (1918)
(f) (g)
FIG.8. Characterizationofindividualpainters. (a,b)GrowthinSofMondrian’sandRenoir’spainting,respectively,overtheirnormalized
careers. Theslopeaofthelinearfit(dashedredlines)is+0.62forMondrianand−0.10forRenoir. (c)Thehistogramofthelinearslopesof
S of1,326modernartistswhoproducedpaintingsinatleastfivedistinctyears. Afewnotableartistsareindicated. Thedashedlineindicates
theaverageslope(a¯ =0.02). (d,e)SamplepaintingsofMondrianandRenoir,respectively,showingtheirstylisticchangesovertheircareers.
(f)Singularityofpaintingsbysevenselectartists. Thedarkerbandindicates−1≤z≤1. (g)Histogramofthesingularityof330artistswith
morethan40paintings.Thedashedlineindicatestheaverageslope(v¯=0.02).
showsthehistogramof330artistswhopaintedmorethan40 negativesingularity(ν = −0.91). EugèneLeroy(1910–2000)
works. In accordance with our definition, we indeed iden- is ranked second in negative singularity, known for numer-
tifythoseknownforahighlevelofsingularityandoriginality ous works featuring obsessively thick brush strokes in dif-
(seeFig.9(b)for100modernartistswhoseνisrankedinthe ferent colors, resulting in obscure objects not readily identi-
top 50, both positive and negative). Qi Baishi (1864–1957), fiable [38]. These findings show that our understanding of
Chinese-bornbutwidelyknownintheWestforhiswittywa- colorcontrastcanindeedcharacterizetheindividualpainters,
tercolorworksofvividcolors[37]showsthehighestpositive and identify those prominently noted for their creativity and
singularity (ν = 0.92). Max Bill (1908–1994) is also highly uniqueness.
singular, known for geometric paintings that also became a
signature of his style as a Swiss designer (ν = 0.79). Kolo-
manMoser(1868–1918),foundingmemberoftheViennaSe-
cession movement and known for repetitive complex motifs
inspired by classical Greek and Roman art, has the largest
9
(a) (b)
... ...
FIG.9. Individualcharacteristicsofmodernartists. (a)The100artistswiththegreatestmetamorphosalityµ. (b)The100artistswiththe
greatestsingularityν.
CONCLUSIONS thoseartistswhoexhibitedvariabilityandoriginality. Wein-
spectedwhetherourmeasurewassensiblebycross-validating
ourfindingswithacceptedunderstandingsoftheirartworks.
Artandculturearethemanifestationsofhumancreativity.
Forthatreason,inadditiontobeingobjectsofappreciationfor Lessonsfromourinvestigationsuggestmanyinterestingdi-
purelyaestheticpurposes,theymaycontainvaluableinforma- rectionsforunderstandingartandcultureviatheuseofmas-
tion we could utilize to understand the creative process. To sivedatasets. Forinstance,Asian,Hindu,andIslamicpaint-
thisend,wehavefocusedonperhapsthemostessentialingre- ing art have been largely untouched in our work; large-scale
dientsofapainting—colorandgeometry—viacolorcontrast analyses of these subjects would be of immediate, universal
and inhomogeneity, which allowed us to quantitatively char- interest. Also,integratingananalyticalstudyusingstylomet-
acterizeandtraceartisticstylesofvariousperiodsandidentify ric measures such as ours with object-recognition and clas-
10
sification techniques from machine learning could lead to a ACKNOWLEDGMENTS
deeper understanding of art that incorporates both the styles
andcontentsofpaintings[24,39].Forexample,howthesame B.L.,D.K.,andH.J.acknowledgethesupportofNational
objects or motifs have been portrayed differently over time ResearchFoundationofKorea(NRF-2011-0028908).J.P.ac-
wouldshedlightonchangesintastesaswellasstyle. Weex- knowledges the support of National Research Foundation of
pectsuchworktofinduseinunderstandingvariousartforms Korea (NRF-20100004910 and NRF-2013S1A3A2055285),
suchassculpture,architecture,visualdesign,film,animation, MinistryofScience(MSIP-R0184-15-1037),andBK21Plus
typography,etc. PostgraduateOrganizationforContentScience.
[1] Gombrich EH. The Story of Art (Vol. 12). London: Phaidon; [19] LyuS,RockmoreD,FaridH.Adigitaltechniqueforartauthen-
1995. tication. Proc. Nat. Acad. Sci. U.S.A. 2004; 101(49):17006-
[2] Manovich L. Data Science and Digital Art History. In- 17010.doi:10.1073/pnas.0406398101
ternational Journal for Digital Art History. 2015; 1. doi: [20] HughesJM,GrahamDJ,RockmoreDN.Quantificationofartis-
10.11588/dah.2015.1.21631 ticstylethroughsparsecodinganalysisinthedrawingsofPieter
[3] KRén E, Marx D. Web Gallery of Art. Available from: Bruegel the Elder. Proc. Natl. Acad. Sci. U.S.A. 2010; 107:
http://www.wga.hu/.(Dateofaccess:2015-10-12). 1279-1283.doi:10.1073/pnas.0910530107
[4] WikiArt.org - the encyclopedia of painting. Available from: [21] KimD,SonSW,JeongH.Large-scalequantitativeanalysisof
http://www.wikiart.org.(Dateofaccess:2015-10-12). paintingarts.Sci.Rep.2014;4:7370.doi:10.1038/srep07370
[5] BBC - Your Paintings. Available from: [22] Graham DJ, Friedenberg JD, Rockmore DN. Field DJ. Map-
http://www.bbc.co.uk/arts/yourpaintings/. (Date of access: pingthesimilarityspaceofpaintings: imagestatisticsandvi-
2015-10-12). sual perception. Visual Cognition, 2010; 18(4):559-573. doi:
[6] DigitalPublicLibraryofAmerica.Availablefrom:http://dp.la/. 10.1080/13506280902934454
(Dateofaccess:2016-04-07). [23] Elgammal A, Saleh B. Quantifying creativity in art networks.
[7] Proctor N. The Google Art Project: A New Generation Sixth International Conference on Computational Creativity
of Museums on the Web?. Curator. 2015; 54(2):215-221. (ICCC)June29-July22015,ParkCity,Utah,USA.
doi:10.1111/j.2151-6952.2011.00083.x [24] De La Rosa J, Sua´rez A. quantitative approach to beauty.
[8] LutosławskiW.TheOriginandGrowthofPlato’sLogic.Long- Perceived attractiveness of human faces in world painting.
mans,Green,London;1897. International Journal for Digital Art History. 2015; 1. doi:
[9] Holmes DI, Kardos J, Who was the author? An in- 10.11588/dah.2015.1.21640
troduction to stylometry. Chance. 2003; 16(2):5-8. doi: [25] Newton I. Opticks, or a treatise of the reflections, refrations,
10.1080/09332480.2003.10554842 inflections&coloursoflight.CourierCorporation;1979.
[10] BinongoJNG,Whowrotethe15thbookofOz?Anapplication [26] GoetheV,EastlakeC.Theoryofcolours.MitPress;1970.
ofmultivariateanalysistoauthorshipattribution.Chance.2003; [27] Ribe N. Steinle F. Exploratory experimentation: Goethe,
16(2):9-17.doi:10.1080/09332480.2003.10554843 Land, and color theory. Phys. Today. 2002; 55:43-50. doi:
[11] Juola P. Authorship attribution. Foundations and 10.1063/1.1506750
Trends in Information Retrieval. 2006; 1:233-334. doi: [28] LandEH.Colorvisionandthenaturalimage.Proc.Natl.Acad.
10.1561/1500000005 Sci.U.S.A.1959;45:636.
[12] Hughes JM, Foti NJ, Krakauer DC, Rockmore DN. Quantita- [29] Land EH. The retinex theory of color vision. Sci. Am. 1977;
tivepatternsofstylisticinfluenceintheevolutionofliterature. 237:108-128.
Proc. Natl. Acad. Sci. U.S.A. 2012; 109(20):7682-7686. doi: [30] BernsR.BillmeyerandSaltzman’sPrinciplesofColorTeech-
10.1073/pnas.1115407109 nology.NY:JohonWiley&Sonsed.3;2000.
[13] ManarisB,RomeroJ,MachadoP,KrehbielD,HirzelT,Pharr [31] GohKI.BarabásiAL.Burstinessandmemoryincomplexsys-
W, Davis RB. Zipf’s law, music classification and aesthetics. tems. Europhys. Lett. 2008; 81(4): 48002. doi:10.1209/0295-
Comput.MusicJ.2005;29(1):55-69. 5075/81/48002
[14] Huron D. The ramp archetype: A study of musical dynamics [32] CharityM.Whatcolorisablackbody?-somepixelrgbvalues.
in 14 piano composers. Psychol. Music. 1991; 19:33-45. doi: Available from: http://vendian.org/mncharity/dir3/blackbody/.
10.1177/0305735691191003 (Dateofaccess:2016-03-15).
[15] Casey M, Rhodes C, Slaney M. Analysis of minimum dis- [33] KleinerF.Gardner’sartthroughtheages:Thewesternperspec-
tances in high dimensional music spaces. IEEE/ACM Trans. tive(Vol.2).CengageLearning;2016.
Audio, Speech, Language Process. 2008; 16:1015-1028. doi: [34] EisenmanSF.CrowTE.LuckacherB.Nineteenthcenturyart:
10.1109/TASL.2008.925883 acriticalhistory.JSTOR;1994.
[16] Sapp C. Hybrid numeric/rank similarity metrics for musical [35] Bomford D. Kirby J. Leighton J. Roy A. Art in the making:
performances.Proc.ISMIR.2008;99:501-506. Impressionism.London:NationalGallery;1990.
[17] TaylorRP,MicolichAP,JonasD.FractalanalysisofPollock’s [36] Howard Mehring - Peyton Wright Gallery. Available from:
drippaintings.Nature.1999;399:422.doi:10.1038/20833 http://peytonwright.com/modern/artists/howard-mehring/.
[18] Taylor RP, Guzman R, Martin TP, Hall GDR, Micolich AP, (Dateofaccess:2016-07-07).
JonasD,ScannelBC,FairbanksMS,MarlowCA.Authenticat- [37] China Online Museum. Available from:
ingPollockpaintingsusingfractalgeometry.2007;28(6):695- http://www.chinaonlinemuseum.com/painting-qi-baishi.php.
702.doi:10.1016/j.patrec.2006.08.012 (Dateofaccess:2016-04-13).