Oxford Handbooks Online - The University of Sydney

The perceptual representation of transparency, lightness, and gloss
OxfordHandbooksOnline
Theperceptualrepresentationoftransparency,lightness,andgloss BartonL.Anderson
OxfordHandbookofPerceptualOrganization
EditedbyJohanWagemans
PrintPublicationDate: Jul2015
OnlinePublicationDate: Aug
2014
Subject: Psychology,Neuropsychology
DOI: 10.1093/oxfordhb/9780199686858.013.030
AbstractandKeywords
Thestructureinlightiscreatedbyitsinteractionwithsurfacesandmaterialsthatfillourenvironment.We
experienceourenvironmentasa3Dlayoutofsurfacesandmaterialsthatpossessqualitiessuchascolor,
lightness,opacity,gloss,andshape.Oneofthemostfundamentalproblemsinvisionscienceinvolves
understandinghowthevisualsystemseparatesthesedifferentsourcesofimagestructureandgeneratesour
experienceofthesedifferentsurfacesandmaterialattributes.Inthischapter,Idiscussthetheoreticalissuesthat
ariseinattemptstocraftsolutionstotheseproblems,anddescribeabodyofempiricaldatathatbearsonthe
relationshipbetweenthedimensionsofperceptualexperienceandthephysicalsourcesofvariabilitygeneratedby
theworld.
Keywords:light,color,lightness,opacity,gloss,shape,visionscience,structure,dimension,perception
1.Theoreticalpreliminaries
Theadaptiveroleofvisionistoprovideinformationaboutthebehaviorallyrelevantpropertiesofourvisual
environment.Ourevolutionarysuccessreliesonrecoveringsufficientinformationabouttheworldtofulfillour
biologicalandreproductiveneedswhileavoidingenvironmentaldangers.Theattempttounderstandvisionasa
collectionofadaptationstospecificcomputationalproblemshasshapedagrowingbodyofresearchthattreats
visionasadecomposablecollectionof‘recovery’problems.Inthisview,perceptualoutputsareunderstoodas
approximatelyidealsolutionstospecificrecoveryproblems,whichhavebeendubbedthe‘naturaltasks’ofvision
(GeislerandRingach2009).Fromthisperspective,thescienceofunderstandingvisualprocessingproceedsby
identifyinganorganism’snaturaltasks,evaluatingtheinformationavailabletoperformeachtask,developing
modelsofhowtoperformataskoptimally,anddiscoveringthemechanismsthatimplementthesesolutions.
Thefirstaspectofthismethodofapproach—theidentificationof‘naturaltasks’—isarguablythemostimportant
becauseitdefinestheproblemthatneedstobesolved.Itisalsotheleastconstrained.Anyenvironmentalproperty
canbehypothesizedtobesomethingthatcouldhaveadaptivevalueandthereforesomethingthatmightprovidea
selectiveadvantagetoanyoneequippedtorecoverit.Presumably,however,onlysomeaspectsofour
environmentwereinvolvedindirectlyshapingtheevolutionofoursenses.Thescientificchallengeisto
differentiatepropertiesthatactuallyexertedselectivepressureinshapingthedesignofoursensesfromthosethat
merelycamealongforthe‘evolutionaryride’(perceptual‘spandrels’).Butthereiscurrentlynoprincipledmeansof
makingsuchdistinctions.Forexample,ageneralargumentcouldbe(andhasbeen)madethatthecomputationof
surfacelightnesswouldbeusefulbecauseitprovidesinformationaboutanintrinsicpropertyoftheexternalworld,
butitismuchhardertofashionaclearargumentabouthowtherecoveryofsurfacealbedoprovidesaspecific
adaptivebenefit,orthatanysuchbenefitplayedaroleinnaturalselection.
Thesecondaspectoftheadaptationistapproach—identifyingtheinformationavailableforacomputation—isin
Page 1 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
principlemoreconstrained.Naturalscenesarerepletewithinformationthatcouldbeusedtosenseaparticular
worldproperty.Oncearecoveryproblemhasbeenidentified,itispossibletoinventorythesourcesofinformation
thatexistinthenaturalworldthatcanbeusedtosenseit.However,mostrecoveryproblemsinvision(suchas
shape,depth,color,lightness,etc.)areconsideredinisolation,oftenininformationallyimpoverishedlaboratory
settings.Thisapproachhasledtothenearlyuniversalacceptanceofabeliefinthepovertyofthestimulus:the
presumptionthattheimagesdonotcontainsufficientinformationtorecovertheaspectsoftheworldthatwe
experience.Thisviewistypicallydefendedbydemonstratingthatitisimpossibletoderiveauniquesolutionfora
specificrecoveryproblembasedontheinformationavailableintheimages.Perceptionisconstruedastheoutputs
ofacollectionofunder-constrainedproblemsofprobabilisticinference,whicharesolvedwiththeaidofadditional
information,assumptions,orconstraints.Soconstrued,itisnaturaltoturntoprobabilitytheoryforguidanceinto
howtosolvesuchinferenceproblemsideally,whichtypicallyentailstheapplicationofBayes’theorem(see
Feldman’schapter,thisvolume).
Thethirdaspectoftheadaptationistprogramisostensiblytheeasiest,andiswheretheorymeetsdata.Perceptsor
perceptualperformanceofobserversiscomparedtothatoftheBayesianideal,constructedonasetofpriorsand
likelihoods.WhendataandtheBayesianidealaredeemedsufficientlysimilar,theexplanatorycircleisconsidered
closed:thefitbetweenmodelanddataisupheldasevidentialsupportforthespecificationofthenaturaltasks,the
selectionofpriorsandlikelihoodsneededtoperformtheinference,andtheclaimthatperceptioninstantiatesa
formofBayesianinference.Allthatremainsisthediscoveryofthemechanismsthatinstantiatesuchcomputations.
Theprecedingdescribeswhatmaycurrentlybeconsideredone(ifnotthe)dominantviewonhowtoapproachthe
studyandmodelingofvisualprocesses.Myownviewdepartsinanumberofsignificantwaysfromthisapproach,
whichshapesbothmyselectionofproblemsandthetheoreticalapproachtakentoaccountfordata.Oneofthe
maingoalsofthischapteristoprovideanoverviewofhowmyapproachhasshapedworkinthreeareasof
surfaceandmaterialperception:transparency,lightness,andgloss.Thegistofmyapproachmaybearticulated
asfollows.First,Iassumethattheattempttoidentifythe‘naturaltasks’ofvision—i.e.,thecomputational‘problems’
thatvisualsystemsputativelyevolvedtosolve—isatbestaguessinggame,andatworstatheoreticalfiction.
Someofthe‘problems’ourvisualsystemsseemtosolvemaybeepiphenomenaloutputs,notexplicitadaptations.
Second,theclaimthatvisionisanill-posedinferenceproblemisalogicalconsequenceoftreatingvisionasa
collectionofrecoveryproblems,forwhichitcanbeshownthatthereisnoclosedformsolutionthatcanbederived
fromtheinformationthatiscurrentlyavailable.Butiftheputative‘recoveryproblem’ismisidentified,orthe
‘informationavailableforsolvingit’isartificiallyrestricted(suchastypicallyoccursinlaboratoryenvironments),
thenitmaynotbevisionthatisill-posed,butourparticularunderstandingofvisualprocessingthatis
misconstrued.
Analternativeapproachistobeginwithwhatwevisuallyexperienceabouttheworld,andattempttodetermine
whatimagepropertiesmodulatetheseexperiences.Thequestionisnotwhetherthereissufficientinformationin
theimagestospecifythetruestatesoftheworld,butrather,whetherthereissufficientinformationtoexplainwhat
weexperienceabouttheworld.Thisapproachisneutralastothe‘computationalgoals’ofthevisualsystem,orif
evenwhethertheideaofacomputationalgoalhasanyrealmeaningforbiologicalsystems.Whereastherecovery
ofaworldpropertycanbeshowntobeunder-constrainedbyargument,thequestionwhetherthereissufficient
informationavailabletoexplainwhatweexperienceabouttheworldisanempiricalquestion.
2.Disentanglingimagesintocausalsources
Weexperiencetheworldasacollectionof3Dobjects,surfaces,andmaterialsthatpossessavarietyofdifferent
phenomenologicalqualities.Thereflectanceandtransmittancepropertiesofamaterial,togetherwithits3D
geometry,structurelightinwaysthatmodulatesourexperienceofshape,lightness,color,gloss,texture,and
translucency.Someimagestructurealsoarisesfromtheidiosyncraticdistributionoflightsourcesinascene—the
illuminationfield.Toafirstapproximation,thislistofsurfaceandmaterialpropertiestendtobeexperiencedas
separatesourcesofimagestructure,despitethefactthattheyareconflatedintheimage.Muchresearchinto
perceptualorganizationhasfocusedonhowthevisualsystemfillsinmissinginformationorgroupsimage
fragmentsintoaglobalstructureorpattern.Whilesuchphenomenaareanextremelyimportantaspectofour
visualexperience,oneoftheotherfundamentalorganizationalproblemsinvolvesunderstandinghowthevisual
systemdisentanglesdifferentsourcesofimagestructureintothedistinctsurfaceandmaterialqualitiesthatwe
Page 2 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
experience.Inwhatfollows,Iconsideravarietyofsegmentationproblemsintheperceptionofsurfaceandmaterial
attributes,andtheinsightsthatsuchproblemsshedonthebroadertheoreticalissuesraisedabove.
2.1.Transparency
Oneoftheperceptuallymostexplicitandtheoreticallychallengingformsofimagesegmentationoccursinthe
perceptionoftransparency.Historically,thestudyoftransparencyfocusedonachromaticsurfaces,whichwas
largelyduetheseminalinfluenceofMetelli’smodeloftransparency(Metelli1970,1974a,1974b,1985;seealso
Gerbino’schapter,thisvolume).Theperceptionof(achromatic)transparentsurfacesgeneratestwodistinct
impressions:itsperceivedlightnessanditsperceivedopacityor‘hidingpower’.Metelli’smodelwasbasedona
simplephysicaldeviceknownasanepiscotister:arapidlyrotatingdiscwithamissingsector.Theproportionof
thediskthatis‘missing’determinestheamountoflighttransmittedfromtheunderlyingsurfacesthroughthe
episcotisterblades,whichisthephysicalcorrelateofatransparentsurface’stransmittance.Thelightness(or
albedo)ofthetransparentsurfacecorrespondedtothecolorofthepaintusedonthefrontsurfaceofthe
episcotister,whichdeterminesthecolorofthetransparentlayer(orforachromaticpaints,itslightness).Metelli’s
modelwasrestrictedto‘balanced’transparency,whichreferredtoconditionswheretheepiscotisterhadauniform
reflectanceandtransmittance,reducingeachtoasinglescalar(number).ForthesimplebipartitefieldsMetelliused
asbackgrounds,thisallowedequationsforthetotalreflectedlightintheregionsofoverlaytobewrittenasasum
oftwocomponents:amultiplicativetransmittanceterm,whichdeterminedtheweightforthecontributionofthe
underlyingsurface;andanadditiveterm,whichcorrespondsthelightreflectedbytheepiscotistersurface.By
construction,Metelliconsidereddisplayscontainingtwouniformlycoloredbackgroundregions,whichgavehima
systemoftwoequationsandtwounknownsthatcouldbesolvedinclosedform.Asignificantbodyofworkshowed
thattheperceptionoftransparencyisoftenwellpredictedbyMetelli’sepiscotistermodel:balancedtransparency
isperceivedwhendisplayswereconsistentwiththeepiscotisterequations,butgenerallynototherwise.Notethat
Metelli’smodelserveddoubledutyasbothaphysicalmodeloftransparencyandapsychologicalmodelofthe
conditionsthatelicitperceptsoftransparency.
Despitethesesuccesses,Metellihimselfnotedacuriousdiscrepancybetweenthepredictionsoftheepiscotister
modelandperception:alightepiscotisterlookslesstransmissivethandarkepiscotister(Metelli1974a).Froma
‘recovery’pointofview,thisconstitutesaperceptualerror,andhencenon-idealperformance,butalmostno
experimentalworkwasconductedtounderstandthisdeviationfromthepredictionsofMetelli’smodel.Wetherefore
performedaseriesofexperimentstotestwhetherthephysicalindependenceofopacityandlightnessisobserved
psychophysically(SinghandAnderson2002).Observersmatchedthetransmittanceofsimulatedsurfacesthat
variedinlightness,andthelightnessoftransparentfiltersthatvariedintransmittance.Wefoundthatlightness
judgmentsweremodulatedbysimulatedtransmittance,andtransmittancejudgmentsweremodulatedbysimulated
variationsinlightness.Thus,althoughthetransmittanceandreflectanceoftransparentlayersarephysically
independentparametersinMetelli’smodel,theyarenotexperiencedasbeingindependentperceptually.
Whattheoreticalconclusionscanbedrawnfromtheseresults?Metelli’smodeltreatedaphysicalmodelof
transparencyasaperceptualmodeloftransparency.Ourfindingsofmutual‘contamination’ofthetransmittance
andlightnessofthetransparentfilterimpliesoneoftwopossibilities:(1)thereisnosimplecorrespondence
betweenthedimensionsofaphysicalmodelandaperceptualmodel,or(2)thatMetelli’smodelisthewrong
physicalmodelonwhichtobasetheoriesofperceivedtransparency.Withrespectto(1),Metelli’smodelequates
theperceivedopacityofanepiscotisterwithitsphysicaltransmittance,andhencecannotexplainwhylight
episcotisterslookmoreopaquethandarkepiscotisters.Thedependenceofperceivedopacityonlightnesscanbe
readilyunderstood,however,ifthevisualsystemreliedonimagecontrasttoassessthehidingpowerof
transparentsurfaces.Alightepiscotisterreducesthecontrastofunderlyingsurfacestructuremorethanan
otherwiseidenticaldarkepiscotister,andhence,shouldappearmoreopaqueifthevisualsystemusesimage
contrasttoassessperceivedopacity1.Indeed,itseemsalmostinevitablethatthevisualsystemutilizescontrastto
judgetheperceivedopacityoftransparentfilters,sincecontrastdeterminesthevisibilityofimagestructurein
general.Butthisimpliesthatthevisualsystemisusingthe‘wrong’imagepropertiestogenerateourexperienceof
aworldproperty,andhencewillalmostalwaysresultinthe‘wrong’answer.Fromtheperspectiveofexplainingour
experience,suchissuesarelargelyirrelevant;theonlyissueiswhetherthereissufficientinformationintheimage
toexplainwhatitisweexperienceabouttheworld,notwhethersuchperceptsareveridical.
Page 3 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
Alternatively,itcouldbe(andhasbeen)arguedthatthediscrepancybetweenperceptionandMetelli’smodel
merelyprovidesevidencethatthereissomethingwrongwithMetelli’smodel,anddoesnotimpactonthemore
generalclaimthatperceptioncanbeidentifiedwiththerecoveryofsomephysicalmodel.FaulandEkroll(2011)
havemadepreciselythisargument.Theycontendthatasubtractivefiltermodelbettercapturestheperceptionof
chromatictransparency,andhencemaybeamoreappropriatemodelofachromatictransparencyaswell.
Althoughthereiscurrentlyinsufficientdatatodeterminewhichofthesealternativesisultimatelycorrectfor
achromaticstimuli,FaulandEkrollreportedsubstantialdiscrepanciesbetweentheirfiltermodelandperceived
transparencywhenthechromaticcontentoftheilluminantwasvaried,despitedemonstratingthattherewas
theoreticallysufficientinformationforamuchbetterlevelofperformance(FaulandEkroll2012).Atthisjuncture,
thereiscurrentlynophysicalmodelthatmapsdirectlyontoourexperienceoftransparentsurfaces,anditis
largelyamatterofscientificfaiththatsuchamodelmayultimatelybediscovered.
2.2.Lightness
Theperceptionoflightnessalsohasbeentreatedasakindofsegmentationproblem.Forachromaticsurfaces,the
termlightness(oralbedo)referstoasurface’sdiffusereflectance.Thelightreturnedtotheeyeisaconflated
mixtureoftheilluminant,surfacereflectance,and3Dpose.Thereiscurrentlyextensivedebateoverthe
computations,mechanisms,and/orassumptionsthatareresponsibleforgeneratingourexperienceoflightness
(seeGilchrist’schapter,thisvolume).Therearefourgeneraltheoreticalapproachestotheproblemoflightness:
scission(orlayersmodels),equivalentilluminantmodels,anchoringmodels,andfilterorfilling-inmodels.Iconsider
eachmodelclassinturn.
2.2.1.Modelsandtheoriesoflightness
Scissionmodels
Scissionmodelsassertthatthevisualsystemderiveslightnessbyexplicitlysegmentingtheilluminantfromsurface
reflectanceinamanneranalogoustothedecompositionthatoccursinconditionsoftransparency.Suchmodels
havebeendubbedlayers,scission,orintrinsicimagemodels(Adelson1999;Anderson1997;Andersonand
Winawer,2005,2008;Barrowetal.1978;Gilchrist1979).Inmodelsoflightness,scissionmodelsassertthatthe
visualsystemteasesapartthecontributionsofreflectance,theilluminant,and3Dpose.Althoughsomeauthors
associatescission(orintrinsicimage)modelswithveridicalperception(Gilchristetal.1999),thereisnothing
inherentinscissionmodelsthatmandatesthisassociation.Theconceptofscissionentailsaclaimabouta
particularrepresentationalformatorprocessofimagedecompositionthatispresumedtounderlieourexperience
oflightness.Thehypothesizedsegmentationprocessesresponsibleforgeneratingtheputativelayered
representationmayormaynotresultinveridicallightnessperceptsdependingonhow(andhowwell)thevisual
systemperformsthehypothesizeddecomposition.
Equivalentillumination
Onemodelthatisconceptuallyrelatedtolayersmodelsistheequivalentilluminationmodel(EIM)developedby
BrainardandMaloney(2011).Aswithlayersmodels,theEIMassumesthatthevisualsystemrecoverssurface
reflectancebyfactoringtheimageintotwocomponents:anestimateoftheilluminant(whichtheyterman
‘equivalentilluminant’)andsurfacereflectance.Whereaslayersmodelshaveassumedthatthereisanexplicit
representationofboththeilluminantandsurfacereflectance,thesameisnotnecessarilytruefortheEIM.TheEIM
isatwo-stagemodelwhichassertsthatthevisualsystembeginsbygeneratinganestimateoftheilluminant,and
usesthisinformationinasecondstagetoderivesurfacereflectancepropertiesfromtheimagedata.Thismodel
remainsmuteastohowthevisualsystemestimatestheparametersoftheestimatedilluminantfromimagesand
alsoremainsuncommittedastotheanyrepresentationalformattheEImaytake.Themainexperimentally
assessableclaimisthatitpredictsthattheparametricstructureofcolororlightnessmatchescanbedescribedby
someEIM.TheapproachoftheEIMcanbeunderstoodasfollows:Givenasetofreflectancematches,isitpossible
tofindamodeloftheilluminantthatisconsistentwiththematches?Notethatthereisnopresumptionthatthe
particularEIMthatputativelyshapesobserver’smatchesisveridical;theonlyclaimisthatobservers’lightness
matchesareshapedbysomeEIM.Indeed,thebenefitofthisclassofmodelisthatitcaninprincipleaccountfor
bothveridicalmatchesand/orthespecificpatternoffailuresinveridicality.
Page 4 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
Anchoringtheory
Athirdtheoreticalapproachtolightnessiscapturedbyanchoringtheory,whichwasdevelopedinanattemptto
accountforavarietyofsystematicerrorsintheperceptionoflightness(Gilchristetal.1999).UnlikelayersorEIM
models,thereisnoexplicitfactorizationoftheilluminantandreflectanceinanchoringtheory.Rather,anchoring
theoryassertsthatperceivedlightnessisderivedthroughasetofheuristicrulesthatthevisualsystemusesto
mapluminanceontoperceivedlightness.Therearetwomaincomponentstoanchoringtheory(seeGilchrist’s
chapter,thisvolume).First,followingWallach(1948),luminanceratiosareusedtoderiveinformationaboutrelative
lightness.Whenthefull30:1rangeofphysicallyrealizablereflectancesarepresentinacommonilluminant,the
truereflectanceofsurfacescanbederivedonthebasisoftheseratiosalone.However,inscenescontainingless
thanthisfull30:1range,someadditionalinformationorruleisneededtotransformambiguousinformationabout
relativelightnessintoanestimateofabsolutesurfacereflectance.Forexample,animagecontaininga2:1rangeof
luminancescouldbegeneratedbysurfaceswithreflectancesofthreepercentandsixpercent,orfivepercent
and10percent,40percent,80,adinfinitum.Anchoringtheoryassertsthatthisambiguitymustberesolvedwith
ananchoringrule,suchthataspecificrelativeimageluminance(suchasthehighest)ismappedontoafixed
lightnessvalue(suchaswhite).Allotherlightnessvaluesinasceneareputativelyderivedbycomputingratios
relativetothisanchorvalue.Anumberoffixedpointsarepossible(e.g.,theaverageluminancecouldbegrey,the
highestluminancecouldbewhite,orthelowestluminancecouldbeblack),butavarietyofexperiments,especially
thosefromGilchrist’slab,havesuggestedthatinmanycontexts,thehighestluminanceisperceivedaswhite.
Filteringandfilling-inmodels
Athirdapproachtolightnesstreatlightnessperceptsastheoutputsoflocalimagefiltersapplieddirectlytothe
images(BlakesleeandMcCourt2004;DakinandBex2003;KingdomandMoulden1988,1992;ShapiroandLu
2011).Suchapproachestypicallydonotdistinguishbetweenperceivedlightness(perceivedsurfacereflectance)
andbrightness(perceivedluminance),atleastnotexplicitlyintheconstructionofthemodel.Rather,anewimage
isgeneratedfromasetoftransformationsappliedtotheinputimage.Inastrictsense,filtermodelsarenottruly
lightnessmodels,sincetheysimplytransformoneimageintoanotherimage.Suchmodelsaremoreappropriately
construedasmodelsofbrightnessthanlightness,sincethereisnoexplicitattempttorepresentsurface
reflectance,ordistinguishreflectancefromluminance.Theirrelevancetounderstandinglightnessdependsonthe
extenttowhichthedistinctionbetweenbrightnessandlightnessmakesbiologicalorpsychologicalsensefora
givenimageorexperimentalprocedure.Likeanchoringmodels,filterapproachestolightnessdonotexplicitly
segmentimageluminanceintoseparatecomponentsofreflectanceandillumination.
Inarelatedmanner,avarietyoffilling-inmodelshavebeenproposedthatdonotexplicitlydistinguishlightness
andbrightness(GrossbergandMingolla1985;ParadisoandNakayama1991;RuddandArrington2001).Such
modelsinvokeatwostageprocess:onethatrespondstothemagnitudeandorientationof‘edges’(oriented
contrast)and/orgradients,andasecondprocessthatpropagatesinformationbetweensuchlocalized‘edge’
responsestogenerateafully‘filled-in’orinterpolatedperceptofbrightnessorcolor.
2.2.2.Evaluatingtheoriesoflightness
Asnotedinarecentarticle,thetopicoflightnessandbrightnesshashistoricallybeenquitedivisive(Kingdom
2011).Onesourceofdisagreementinvolvestheverydistinctionbetweenbrightnessandlightness.Althoughsuch
constructsareeasilydistinguishedfromeachotherwithregardtotheirintendedphysicalreferents,itisnotclear
that(orwhen)suchdistinctionshavepsychologicalmeaning.Thedistinctionbetweenlightnessandbrightnessis
particularlyproblematicforthekindsofdisplaysthataretypicallystudiedineitherlightnessorbrightnessstudies.
Inalmostallcases,thetargetsofinteresthaveasingle,uniformluminance(orapproximatelyso),andare
embeddedinhighlysimplifiedgeometricandilluminationcontexts.Forscenesdepictingrealorsimulatedsurfaces,
thesurfacesofinterestaretypicallyflat,matte,andarrangedinasingledepthand/orilluminant.Theytypically
lackinformationaboutthelightfield,suchasthatprovidedbyspecularreflections,3Dstructure,shading,and
inter-reflections.Itisperhapsnotsurprising,then,thatthefieldremainsdividedastotheproperwaytounderstand
howsuchimpoverisheddisplaysareexperienced,sinceitisunclearwhetherthedistinctionbetweenlightnessand
brightnessispsychologicallymeaningfulinmanyofthesedisplays.Inwhatfollows,Iwillconsidersomerecent
evidencerelevantforeachofthetheoriesoflightnessdescribedabove.
Page 5 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
Clicktoviewlarger
Figure1 .StereoscopicKanizsafiguredemonstratingtheroleofscissiononperceivedlightnessfortwo
differentgreyvalues.Thesmallpieshapedinducingsectorsarethesameshadeofdarkgreyinthetoptwo
rows,andthesameshadeoflightgreyinthebottomtworows.Whenthelefttwoimagesarecrossfused,
ortherighttwoimagesdivergentlyfused,anillusorydiamondisexperience.Notethatthediamondsinthe
firstandthirdrowsappearmuchlighterthantheircorrespondingfiguresinthesecondandfourthrows.
AdaptedfromAnderson(1998,TrendsinCognitiveScience).
Thecoreclaimofscissionmodelsisthatourexperienceoflightnessinvolvesthedecompositionoftheinputinto
separablecauses.Oneofthedifficultiesinassessingscissionmodelsisthatitisnotalwaysclearwhether(or
when)suchseparationoccurs,orwhatcriteriathatshouldbeappliedtodeterminewhethersuchdecomposition
occurs.Onecanbeginbyposingaquestionofsufficiency:Canscissioninducetransformationsinperceived
lightnesswhenitisphenomenallyapparent?Themostphenomenologicallycompellingsenseofscissionoccursin
conditionsoftransparency,whichrequiresthesatisfactionofbothgeometricandphotometricconditions.One
techniqueforinducingscissioninvolvesmanipulatingtherelativedepthandphotometricrelationshipsof
stereoscopicKanizsafiguressuchasthosedepictedinFigure1.Whenthegrey,wedge-shapedsegmentsofthe
Kanizsafigure’sinducingelementsinFigure1aredecomposedintoatransparentlayeroverlyingawhitedisk
(secondandfourthrowsofFigure1),theyappearsubstantiallydarkerthanwhenthesamegreysegmentappears
tooverlieadarkdisk(firstandthirdrowsofFigure1).Notethatthecoloroftheunderlyingcircularinducing
elementappearstobe‘removed’fromthegreywedge-shapedsegmentsandattributedtothemoredistantlayer,
whichputativelytransformstheperceivedlightnessofthetransparentlayer.Notealsothatthedirectionofthe
lightnesstransformationdependsonwhichlayerobserversareaskedtoreport.Ifobserversareaskedtoreport
thecolorofthefarlayerunderneaththegreysectorsofthetopimage,theyreportitasappearingquitedark
(nearlyblack),sincethisisthecoloroftheinterpolateddisc.Butiftheyareaskedtoreportthenearlayerofthe
transparentregion,theyreportitasappearingquitelight.
Page 6 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
Clicktoviewlarger
Figure2 .Stereoscopicnoisepatternscanalsobedecomposedintolayersinwaysthatinducelarge
transformationsinperceivedlightness.Ifthelefttwoimagesarecrossfusedortherighttwoimages
divergentlyfused,thetopimageappearstosplitintoapatternofdarkcloudsoverlyinglightdiscs(top),or
lightcloudsoverlyingdarkdisks(bottom).Thetexturesinthetopandbottomarephysicallyidentical.
AdaptedfromAnderson(1999).
Clicktoviewlarger
Figure3 .Scissioncanalsobeinducedbyaselectivegroupingthelightanddarkcomponentsoftextureof
thetargets(chesspieces)withthesurround.Thetextureswithinthechesspiecesinthetopandbottom
imagesareidentical,butappearasdarkcloudoverlyinglightchesspiecesonthetop,andlightclouds
overlyingdarkchesspiecesonthebottom.
AdaptedfromAndersonandWinawer(2005).
Inordertoprovidemoreconclusiveevidencefortheeffectsofscissiononperceivedlightness,Iconstructed
stereoscopicvariantsofFigure1usingrandomnoisetextures.Thegoalwastoinducetransparencyinatexture
suchthatthelightanddark‘components’ofthetexturewouldperceptuallysegregateintodifferentdepthplanes.
AnexampleispresentedinFigure2.Whenthelefttwocolumnsarecross-fused,vividperceptsofinhomogeneous
transparencycanbeobserved:Thetopimageappearsasdarkcloudsoverlyinglightdisks,andthebottom
appearsaslightcloudsoverlyingdarkdisks.Notethatthelightestcomponentsofthetextureinthetopimage
appearasportionsoftheunderlyingdiscinplainview,whereasthesameregionsinthebottomimageappearas
themostopaqueregionsofthelightcloudsinthebottomimage(andviceversaforthedarkregions).We
subsequentlyshowedthatsimilarphenomenacouldbeobservedinnon-stereoscopicdisplays.Intheseimages,
scissionwasinducedbyembeddingtargetsinsurroundsthatcontaintexturesthatselectivelygroupwitheitherthe
lightordark‘components’ofthetextureswithinthetargets(Figure3).Aswiththeirstereoscopicanalogues,the
whiteandblackchesspiecesareactuallyphysicallyidentical(i.e.,containidenticalpatternsoftexture).Notethat
Page 7 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
theluminancevariationswithinthetextureofthechesspiecefiguresareexperiencedasvariationsintheopacity
ofatransparentlayerthatoverlieauniformlycoloredsurface.Theopacityofthetransparentsurfaceisgreatest
forluminancevaluesthatmostcloselymatchthesurroundalongthebordersofthechesspieces(darkontop,light
onthebottom),andtheleastopaquewhenforluminancevaluesthataremostdifferentfromthesurround(lighton
top,darkonthebottom).Notethatthelightestregionswithinthetargetsonthedarksurroundappearinplainview,
andthedarkestregionswithinthetargetsappearinplainviewonthelightsurround.Thisbiasisevidentfor
essentiallyallrangesoftargetluminancetested,althoughthisperceptualfactisinnowaymandatedbythe
physicsoftransparency,particularlyforunderlyingsurfacesthatdonotappearblackorwhite.
Thesephenomenademonstratethatscissioncaninducestrikingtransformationsinperceivedlightnessin
conditionsoftransparency,butitdoesnotaddressthebroaderquestionofwhetherscissionplaysarolein
generatingourexperienceoflightnessinconditionsthatdonotgenerateexplicitperceptsofmultiplelayersor
transparency.
EIMsalsoassertthattheperceptionofsurfacecolorandlightnessisderivedbydecomposingtheimageinto
estimatesoftheilluminantandsurfacereflectance.Theevidenceinsupportofthismodelis,however,
phenomenologicallyindirect.WorkfromBrainard’sandMaloney’slabshavedemonstratedthattheparametric
structureofavarietyofmatchingdatacanbeexplainedwithatwo-stagemodelinwhichthefirststageinvolvesan
estimationoftheilluminant(an‘equivalentilluminant’),whichisthenusedtoderiveobservers’reflectancematches
fromtheinputimages(BrainardandMaloney,2011).
UnlikescissionmodelsorEIMs,anchoringtheoryassertsthatlightnessisderivedwithoutexplicitlydecomposing
theimagesintoanexplicitrepresentationofilluminationandreflectance.Thecentralpremiseofanchoringtheory
isthatthevisualsystemsolvestheambiguityoflightnessbytreatingaparticularrelativeluminanceasafixed
(anchor)pointonthelightnessscale(namely,thatthehighestluminanceaswhite),independentofthelevelof
illuminationorabsoluteluminancevaluesinascene.Totestthisclaim,weconstructedbothpaperMondrians
displayedinanotherwiseuniformlyblacklaboratory,andsimulatedMondriansdisplayedonaCRTinadarkblack
labroom(Andersonetal.2008).Inallcases,thehighestluminanceintheroomwasthecentraltargetpatchofthe
Mondriandisplay.Wevariedboththereflectancerangeandilluminationleveloftheformer(i.e.,paperMondrians),
andthesimulatedreflectancerangeandsimulatedilluminantlevelsofthelattersimulatedMondrians.Forrestricted
reflectanceranges(3:1orless),wefoundthatthehighestluminancecouldvaryinperceivedlightnessasa
functionofillumination.ForoursimulatedilluminantsandMondriandisplays,observers’lightnessmatches
(expressedasapercentageofreflectance)werealogarithmicfunctionof(simulated)illuminant,ratherthanan
invariant‘white’aspredictedbyanchoringtheory.Theseresultssuggestthattheapparent‘anchoring’of
luminanceto‘white’isaconsequenceoftheparticularexperimentalconditionsthathavebeenusedtoassessthis
model,ratherthanreflectinganinvariant‘anchorpoint’usedtoscaleotherlightnessvalues.
Somerecentdatahasprovidedsomestrongevidenceagainstanexplicitilluminationestimationmodel,andmore
generally,anymostthatreliesonluminanceratiostocomputeperceivedlightness(suchasanchoringtheory).
Radonjicetal.(2011)conductedexperimentsdepictingcheckerboarddisplaysinadisplaycapableofdisplaying
anextremelylargedynamicrange,andfoundthatobserversmappedaveryhighdynamicrange(~10,000:1)onto
anextendedlightnessrangeof100:1,whichspannedfrom‘white’to‘darkblack’(thedarkestvalueswere
obtainedusingglossypapers).Suchbehaviorwouldnotbeexpectedforanymodelthatattemptstoinfera
physicallyrealizableilluminant,oranyrealizablereflectanceratiosofrealsurfaces,asembracedbyanchoring
theoryortheEIM.
OnecommonassumptionofanchoringtheoryandtheEIMisthatthevisualsystemexplicitlyattemptstoextractan
estimateoflightnessthatcorrespondstothephysicaldimensionofsurfacealbedo.TheresultsofRadonjićetal.
(2011)providecompellingevidenceagainstthisview.Justasourexperienceoftransparencymaynothaveany
directcorrespondencetothephysicaldimensionsthatmodulateperceivedtransparency(suchastransmittance),
theperceptionoflightnessmaynotrepresentanapproximationofthephysicaldimensionofsurfacealbedo.The
resultsofRadonjicetal.provideevidencethatdirectlychallengeanyattempttointerpretthevisualresponseasa
‘bestguess’astotheenvironmentalsourcesthatproducedtheirstimuli,sincethereisnocombinationofsurface
reflectanceandilluminantthatcanproducesuchstimuli(atleastinacommonilluminant).Iwillreturntothis
generalpointinthegeneraldiscussionbelow.
Page 8 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
3.Gloss
Theexperienceofglossisanotheraspectofourexperienceofsurfacereflectancethathasreceivedagrowing
amountofexperimentalattention.Whereastheconceptofsurfacelightnesshasbeencastastheproblemof
understandinghowweexperiencethediffusereflectanceofasurface,theperceptionofglossistypicallycastas
theproblemofunderstandinghowweexperiencethespecular‘component’ofreflectance.Fromagenerativepoint
ofview,thediffuseandspecular‘components’ofreflectancearetreatedascomputationallyseparable.So
construed,theproblemofglossperceptioninvolvesunderstandinghowthevisualsystemsegmentstheimage
structuregeneratedbyspecularreflectancefromdiffusereflectance(andallothersourcesofimagestructure).
Theapparentintractabilityofthisproblemhasinspiredattemptstofindcomputationalshort-cutstoavoidthe
complexityofthisdecompositionproblem.Oneapproachassertsthatthevisualsystemusessimpleimage
statisticsthatdonotrequireanyexplicitdecompositionoftheimagesintodistinctcomponentsofreflectanceto
deriveourexperienceofgloss.Motoyoshietal.(2007)arguedthatperceivedglosswaswellpredictedbyan
image’shistogramorsub-bandskew,ameasureoftheasymmetryofthepixelhistogram(orresponseofcentersurroundfilters)respectively.Thisclaimwasevaluatedforaclassofstuccosurfaceswithastatisticallyfixedlevel
ofsurfacereliefthatwereviewedinfixedilluminationfield.Intheseconditions,glossysurfacesgeneratedimages
withastrongpositiveskew,whereasmattesurfacesgeneratedsurfaceswithnegativeskew.Theattractivefeature
ofthiskindofmodelisthatitpotentiallyreducesacomplexmid-levelvisionproblemintoacomparativelysimple
problemofdetectinglow-levelimageproperties.
Clicktoviewlarger
Clicktoviewlarger
Figure4 .Theperceptionofglossdependscriticallyhighlightsappearinginthe‘rightplaces’ofasurface’s
diffuseshadingprofile.InA,thehighlightsappearneartheluminancemaximaofthediffusedshading
Page 9 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
profileandhavesimilarorientations,andthesurfaceappearsrelativelyglossy.InB,thehighlightshave
beenrotatedsothattheyappearwithrandompositionsandorientationsrelativetothediffuseshading
profile,anddonotappearglossy.
AdaptedfromAndersonandKim(2009).
However,subsequentworkhasshownthatourexperienceofglosscannotbeunderstoodsoeasily(Andersonand
Kim2009;KimandAnderson2010;Kimetal.2011;Marlowetal.2011;OlkkonenandBrainard2010,2011).Oneof
themainproblemswiththeproposedimagestatisticsisthattheyfailtotakeintoaccountthekindofimage
structurethatpredictswhenglosswillorwon’tbeperceived.Specularhighlights,andspecularreflectionsmore
generally,mustappearinthe‘rightplaces’onsurfacestoelicitaperceptofgloss(seeFigure4).Fromaphysical
perspective,specularhighlightsclingtoregionsofhighsurfacecurvature.Theperceptionofglossalsorequires
highlightstoappearinspecificplacesandhaveorientationsconsistentwithsurfaceshadingforasurfaceto
appearglossy,ageometricconstraintthatisnotcapturedbyhistogramorsub-bandskew.
Clicktoviewlarger
Figure5 .Interactionsbetween3Dshapeandperceivedglossasafunctionoftheilluminationfield.Allof
theimagesinthisimagehavethesamephysicalglosslevel,butdonotappearequallyglossy.Theimages
inthetoprowwererenderedinanilluminationfieldwheretheprimarylightsourceswereorientedobliquely
tothesurface,andtheimagesinthesecondrowwereilluminatedinthesameilluminationfieldwiththe
primarylightsourcesorientedtowardsthesurface.
AdaptedfromMarlowetal.(2012).
Clicktoviewlarger
Figure6 .Dataandmodelfitsfortheexperimentsweperformedontheinteractionsbetweenperceived
gloss,3Dshape(ascapturedbyameasureofsurfacerelief),andtheilluminationfield.Thestimuliwere
viewedeitherwithorwithoutstereoscopicdepth(the‘disparity’and‘nodisparity’conditionsrespectively).
Thedifferentcoloredcurvesineachgraphcorrespondtoadifferentilluminationdirectionofaparticular
illuminationfield(called‘Grace’).Theglossjudgmentsareinthetwotoprightpanels.Thepanelsonthe
leftrepresentthejudgmentsofaseparategroupofobserversoffourdifferentcuestogloss:thedepth,
coverage,contrast,andsharpnessofspecularreflections.Thepanellabeled‘skew’wascomputeddirectly
Page 10 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
fromimages.Thedottedlinesinthetwographsonthetoprightcorrespondtothebestfittinglinear
combinationofthecuesontheleft,whichaccountfor94percentofthevarianceofglossjudgments.The
weightsaredenotedintheboxesadjacenttothesmallarrowsinthecenterofthegraphs.
AdaptedfromMarlowetal.(2012).
Althoughtheseresultssuggestthatthevisualsysteminsomesense‘understands’thephysicsofspecular
reflection,thereareotherfindingsthatrevealthattheextentofanysuchunderstandingislimited.Theperception
ofglosshasbeenshowntointeractwithasurface’s3Dshapeanditslightingconditions,whicharephysically
independentsourcesofimagevariability(Hoetal.2008;Marlowetal.2012;OlkkonenandBrainard2011).These
interactionshavebeenobservedbyavarietyofauthorsandhaveresistedexplanation.Indeed,theseinteractions
aredifficulttounderstandfromaphysicalperspective,sinceglossand3Dshapeareindependentsourcesof
imagestructure.However,werecentlypresentedevidencethattheseinteractionscanbeunderstoodasa
consequenceofasimplesetofimagecuesthatthevisualsystemusestogenerateourexperienceofgloss,which
areonlyroughlycorrelatedwithasurface’sphysicalglosslevel(Marlowetal.2012).Someoftheintuitionshaping
thistheoreticalproposalcanbegainedbyconsideringthesurfacesdepictedinFigure5.Allofthesurfacesin
theseimageshavethesamephysicalglosslevel,yetappeartovaryappreciablyinperceivedgloss.Eachcolumn
containssurfaceswithacommondegreeofrelief,andeachrowcontainsimagesthatwereplacedinan
illuminationfieldwiththesamedirectionoftheprimarylightsources.Wevariedthestructureofthelightfield,the
directionoftheprimarylightsources,and3Dsurfacerelief.Observersperformpairedcomparisonjudgmentsof
theperceivedglossofallsurfaces,wheretheychosewhichofapairofsurfaceswasperceivedasglossier.The
datarevealedcomplexinteractionsbetweenthelightfieldandsurfaceshapeonglossjudgments.Ascanbeseen
inFigure6,thevariationoftheilluminationfieldandshapehadasignificantimpactonthesharpness,size,and
contrastofspecularhighlightsintheseimages.Wereasonedthatifobserverswerebasingtheirglossjudgments
onthesecues,thenitshouldbepossibletomodelobservers’glossjudgmentswithaweightedcombinationof
theseimagecues.However,thereiscurrentlynoknownmethodforcomputingthesecuesdirectlyfromimage.We
thereforehadindependentsetsofobserversjudgeeachofthesecues,andtestedwhetheritwaspossibleto
predictglossjudgmentswithaweightedsumofthesecues.Wefoundthatasimpleweightedsummodelwas
capableofpredictingover94percentofthevarianceoftheotherobservers’glossjudgments.Thus,althoughthe
perceptionofsurfaceswiththesamephysicalglosslevelcanappeartovarysignificantlyinperceivedgloss,these
effectscanbeunderstoodwithasetofrelativelysimple,albeitimperfect,‘cues’thatthevisualsystemusesto
generateourexperienceofgloss.
4.Theperceptualorganizationofsurfacesandmaterials
Thelastfewdecadeshavewitnessedanexplosiveincreaseinmodelsthathavetreatedvisualprocessesasa
collectionofapproximatelyideal‘solutions’toparticularcomputationalproblems.Suchmodelsareexplicitly
teleological:theytreatadesiredoutcome,goal,ortaskastheorganizingforcethatshapestheperceptualabilities
theyareattemptingtomodel.Evolutionarytheoryservesastheengineeringforcethatputativelydrivesbiological
systemstowardoptimalsolutions.Thismodelingprocesshingescriticallyontheabilitytospecifythe‘naturaltasks’
thatwereputativelyshapedbyevolution.Thejustificationfortheadaptiveimportanceofaparticular‘naturaltask’
typicallytakesagenericform:anenvironmentalpropertyistreatedashavingevolutionarysignificancebecauseit
isanintrinsicpropertyoftheworld.Thus,anyanimalcapableofaccuratelyrecoveringthatpropertywouldgainan
adaptiveadvantage.Thepropertiestoberecovered—the‘tasks’ofvision—aredefinedinwithrespecttoparticular
physicalsourcesofvariability.Ourexperienceoflightnessistreatedasthevisualsystem’ssolutiontotheproblem
ofrecoveringthealbedoofasurface.Ourexperienceoftransparencyistreatedastheperceptualsolutiontoa
particulargenerativemodeloftransparency(suchasMetelli’sepiscotistermodelorFaulandEkroll’sfiltermodel).
Andourexperienceofglossisunderstoodasthevisualsystem’sattempttoestimatethespecularcomponentof
surfacereflectance.
Oneoftheassumptionsofthisapproachisthatthedimensionsofpsychologicalvariationareassumedtomirror
thesourcesofphysicalvariation.ThisassumptionisexplicitinbothMetelli’smodel,whichtreatedtheepiscotister
asbothaphysicalandpsychologicalmodeloftransparency,andtheEIMofBrainardandMaloney,whichasserts
thatthevisualsystemgeneratesa‘virtual’modeloftheilluminanttorecovercolorandlightness.Theperceptionof
glosshasalsobeenstudiedasakindof‘constancy’problem,whichinvolvesrecoveringthespecular‘component’
ofreflectance.
Page 11 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
Amainthemeofthischapteristoquestiontheadequacyofthisconceptualizationofvision.Ratherthanattempting
toguessthe‘naturaltasks’andananimal,Iviewthegoalofperceptualtheorytodiscoverthe‘natural
decompositions’ofrepresentationalspace,i.e.,todiscoverthepsychologicaldimensionsthatcapturethespace
ofourexperiences.Theprecedingfocusedonourexperienceoftransparency,lightness,andgloss.Eachofthese
attributescanbeidentifiedwithaparticularphysicalpropertyofsurfacesandmaterial,whichcanbedescribedin
physicaltermsindependentlyofanyperceptualsystem.Suchdescriptionsassumethatthevisualsystemplaysno
partindefiningtheattributesthatitputativelyrepresents;thedimensionsaregivenbyidentifiablesourcesof
variationintheworld,whichthevisualsystemisattemptingtorecover,notbyintrinsicpropertiesofthevisual
system.Weareleftdiscussinghowwellthevisualsystemencodesorrecoversaparticularworldproperty,rather
thanhowthevisualsystemcontributestoshapingthedimensionsofourvisionexperience.
Theprecedingsuggeststhatthisgeneralapproachfailstoexplainanumberofdifferentphenomenainsurfaceand
materialperception.TheperceptionofsurfaceopacitydoesnotfollowMetelli’smodeloftransmittance.Weargued
thatoneofthemainreasonsforthisfailurewasthatMetelli’smodelisbasedonaratioofluminancedifferences,
wherearenotavailabletoavisualsystemthattransformsretinalluminanceintolocalcontrastsignals.Weshowed
thatourmatchingdatawerewellpredictedbyamodelinwhichobserversmatchedcontrastratios,ratherthan
luminancedifferenceratios.Oneofthekeypointsofourmodelwastodefinetransmittanceinawaythatwas
consistentwithintrinsiccodingpropertiesofthevisualsystem,evenifthisresultsinthefailuretocompute
physicallyaccuratemeasureofsurfaceopacity.Thisgeneralapproachofaphysiologicallymotivatedmodelhas
alsobeenpursuedbyarecentmodeloftheseresultsbyVladusich,whoproposedanalternativemodelofour
transmittancematchingdata(Vladusich2013).Heshowsthatourtransmittancematchingdatacanbecaptured
withamodifiedversionofMetelli’smodelinwhichlogluminancevaluesareusedinsteadofluminancevalues
(Vladusich,submitted).Likeourmodel,thechoicetouseLogluminancevaluescannotbederivedfromthephysics
oftransparentsurfaces;theyarederivedfromintrinsicresponsepropertiesofthevisualsystem.
Thedifferenttheoriesoflightnessperceptionareevenmorecontentiousanddiversethanthosefoundinthe
transparencyliterature.Oneofthebasicissuesinvolvesthedistinctionbetweenlightnessandbrightness.The
perceptionoflightnessisthendefinedastheperceptionofdiffuse(achromatic)surfacereflectance,whereas
brightnessisdefinedastheperceptionofimageluminance.Thepresumptionisthatthesephysicaldistinctions
havepsychologicalmeaning.Butthisisfarfromself-evident.Themajorityofworkonlightnesshasused2D(flat)
mattedisplaysofsurfaceswithuniformalbedos,forwhichthedistinctionbetweenlightnessandbrightnessis
arguablyleastvalid(ormeaningful)perceptually.Forsomeexperimentalconditions,observers’matchingdatawill
differsubstantiallyifinstructedtomatcheitherbrightnessorlightness.Butinothers,adifferenceininstructions
maymakelittleornodifference.Consider,forexample,theproblemofmatchingthe‘brightness’versusthe
‘lightness’ofthechecker-shadowillusion.Agivenpatchappearsaparticularshadeofgrey,andthereisno
evidencethatobserverscoulddistinguishitsbrightnessandlightness.Insupportofthisview,wefoundthatthe
perceptionoflightnessincreasedasafunctionofitsluminanceinbothsimulatedand‘real’Mondriandisplays.
Moreover,thedataofRadonjićetal.(2011)demonstratethatobserverswillreadilymapaphysicallyunrealizedset
ofluminances,spanning4ordersofmagnitude,ontoalightnessscaletwoorderssmaller.Theseresultsare
impossibletoreconcilewithmodelsthattreattheproblemoflightnessasarecoveryproblem,sincetherangeof
reflectancesinanaturalscenecanonlyspanarangeof~30:1.
Intheperceptionofgloss,wefoundthatobserver’sexperienceofglosscanbewellpredictedbyasetofsimple
cuesthatareonlyimperfectlycorrelatedwiththephysicalglossofasurface.Glossisnotdefinedwithrespectto
somephysicallyspecifieddimensionofsurfaceoptics,butwithrespecttoasetofcuesthevisualsystemusesas
aproxyforanobjectivelydefinedsurfaceproperty.
Whatgeneralunderstandingcanbegleanedfromthesepatternsofresults?Alloftheseresultsrevealthe
insufficiencyofattemptingtoidentifypsychologicaldimensionsofourexperiencewithphysicalsourcesofimage
variability.Thefactthatwehaveaparticularexperienceoflightness,gloss,ortransparencydoesnotimplythat
thedimensionsofourexperiencemapontoaparticularphysicaldimensionand/oritsparameterization.The
generalargumentusedtojustify‘naturaltasks’takesthegenericformthat‘gettinganenvironmentalpropertyright
increasesadaptivefitness.’Thepresumedidentificationoffitnesswithveridicalperceptionisactuallyfallacious
(seeHoffman2009;cf.Lewontin1996),butevenifsuchviewswereaccepted,theyareincapableofdistinguishing
perceptualabilitiesthatwereactuallyshapedbynaturalselectionfromthe‘spandrels’thatcamealongforthe
evolutionaryride.Thefactthathumanobserverswillreadilymapanecologicallyunobtainablerangeofluminance
Page 12 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
values(inasingleilluminant)ontolightnessestimatessuggeststhatlightnessmaybeoneexampleofaperceptual
spandrel.Althoughhumanobserverscanusuallydistinguishreflectancedifferencesfromothersourcesofimage
variation,theperceptionofabsolutelightnessmaysimplybetheresultoflow-levelprocessesofadaptationthat
allowthevisualsystemtoencodeaparticularrangeofluminancevalues.Indeed,Iamawareofnocompelling
evidenceorargumentaboutwhylightnessconstancyperseprovidedanadaptiveadvantage,orissomethingthat
thevisualsystemisexplicitly‘designed’tocompute.Asimilarargumentholdsfortheperceptionoftransparency
andgloss.Wecanreadilydistinguishbetweensurfacesormediathattransmitlightfromthosethatdonot,or
distinguishbetweensurfacesthatreflectlightspecularlyfromthosethatdonot.Butthedataalsosuggeststhatwe
donotscalethesedimensionsinawaythatisphysicallycorrectforanyoftheseproperties.
Althoughitisdifficulttocraftacompellingargumentforthespecificadaptiveutilityofdevelopingaphysically
accuratemodeloflightness,gloss,andtransparency,thefactthatweexperiencethesedifferentsourcesof
variableasdifferentunderlyingcausesimpliesthatthevisualsystemiscapableofatleastqualitatively
distinguishingdifferentsourcesofimagestructure.This‘sourcesegmentation’isarguablyoneofthemost
importantgeneralpropertiesofourvisualsystem.Thevisualsystemmay,infact,bequitepoorinestimating
lightnessinarbitrarycontexts,butitisnonethelesstypicallyquitegoodatdistinguishingimagestructuregenerated
bylightnessdifferencesfromilluminationchanges,orvariationsintheopacityofatransparentsurface,orfrom
specularreflections.Theidentificationofspecularreflectionsasspecularreflectionsdependsontheircompatibility
withdiffusesurfaceshadingand3Dsurfacegeometry,andismodulatedbythestructure,intensity,anddistribution
ofimagestructuresoidentified,evenifitdoesnotaccuratelycapturethe‘true’glosslevelofasurface.And
althoughthephysicaltransmittance(oropacity)ofasurfacedoesnotvaryasafunctionofitsalbedoorcolor,the
psychologicalanalogofopacity—its‘hidingpower’—willforavisualsystemthatusescontrasttodeterminethe
visibilityofimagestructure.Thevisualsystemmaynotdeterminethe‘true’opacityofasurface,butnonethelessis
effectiveatperformingasegmentationthatcapturesthepresenceorabsenceoftransmissivesurfacesandmedia.
5.Summaryandconclusions
Inthischapter,Ihaveconsideredanumberoftopicsintheareaofsurfaceandmaterialperception:transparency,
lightness,andgloss.Theorganizationofthesetopicswaslargelyshapedbymyhistoricalprogressionin
conductingresearchintoeachofthesedomains;manyalternativeorganizationsarepossible.Inalloftheseareas
ofinquiry,therehasbeenastrikingtendencytotreatphysicalmodelsofimageformationassomekindof
approximationtoaperceptualmodeloftheirapprehension.Theprecisewaythataphysicalmodel‘counts’asa
psychologicalmodelistypicallyleftunspecified.Itappearstobebasedonsomeintuitionthatthevisualsystem
‘knows’or‘understands’thephysicsthatofaparticularsurfaceormaterialattribute.Icontendthatoneofthemain
goalsofvisionscienceshouldbetodiscoverthedimensionsofperceptualexperience,andtheimagevariables
thatmodulateourresponsetothem.Whereasthedimensionsofphysicalvariablescanbespecifiedindependently
ofanyperceptualsystem,thedimensionsofperceptualexperienceareinherentlyrelational,andmustconsider
theintrinsicpropertiesofthevisualsystemaswellastheenvironmentsinwhichtheyoperate.
References
Adelson,E.H.(1999).‘Lightnessperceptionandlightnessillusions’.InThenewcognitiveneurosciences,2nded.,
pp.339–51.(Cambridge,MA:MITPress).
Anderson,B.L.(1997).‘Atheoryofillusorylightnessandtransparencyinmonocularandbinocularimages:the
roleofcontourjunctions’.Perception26(4):419–53.
Anderson,B.L.(1998).Stereovision:Beyonddisparitycomputations.TrendsinCognitiveSciences,2:222–228.
Anderson,B.L.(1999).Stereoscopicsurfaceperception.Neuron,24:919–928.
Anderson,B.L.,andKim,J.(2009).‘Imagestatisticsdonotexplaintheperceptionofglossandlightness’.Journalof
Vision9(11):1–17.
Anderson,B.L.,andWinawer,J.(2005).‘Imagesegmentationandlightnessperception’.Nature434(7029):79–83.
Page 13 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
doi:10.1038/nature03271.
Anderson,B.L.,andWinawer,J.(2008).‘Layeredimagerepresentationsandthecomputationofsurfacelightness’.
JournalofVision8(7):18,11–22.doi:10.1167/8.7.18.
Anderson,B.L.,deSilva,C.,andWhitbread,M.(2008).‘Lightnessperceptionhasnoanchor’.JournalofVision
8(6):284.
Barrow,H.G.,Tenenbaum,J.M.,Hanson,A.,andRiseman,R.(1978).‘Recoveringintrinsicscenecharacteristics
fromimages’.ComputerVisionSystems,pp.3–26.(NewYork:AcademicPress).
Blakeslee,B.,andMcCourt,M.E.(2004).‘Aunifiedtheoryofbrightnesscontrastandassimilationincorporating
orientedmultiscalespatialfilteringandcontrastnormalization’.VisionResearch44(21):2483–503.doi:
10.1016/j.visres.200405.015.
Brainard,D.H.,andMaloney,L.T.(2011).‘Surfacecolorperceptionandequivalentilluminationmodels’.Journalof
Vision11(5),doi:10.1167/11.5.1.
Dakin,S.C.,andBex,P.J.(2003).‘Naturalimagestatisticsmediatebrightness“fillingin”’.ProcBiolSci270(1531):
2341–8.doi:10.1098/rspb.2003.2528.
Faul,F.,andEkroll,V.(2011).‘Onthefilterapproachtoperceptualtransparency’.JournalofVision11(7):doi:
10.1167/11.7.7.
Faul,F.,andEkroll,V.(2012).‘Transparentlayerconstancy’.JournalofVision12(12):1–26.doi:
10.1167/12.12.7.
Feldman(thisvolume),InTheHandbookofPerceptualOrganization,editedbyJ.Wagemans.(Oxford:Oxford
UniversityPress).
Geisler,W.S.,andRingach,D.(2009).‘Naturalsystemsanalysis.Introduction’.VisNeurosci26(1):1–3.
Gerbin(thisvolume),InTheHandbookofPerceptualOrganization,editedbyJ.Wagemans.(Oxford:Oxford
UniversityPress).
Gilchrist,A.L.(1979).‘Theperceptionofsurfaceblacksandwhites’.SciAm240(3):112–2,124.
Gilchrist,A.,Kossyfidis,C.,Bonato,F.,Agostini,T.,Cataliotti,J.,Li,X.J.,…Economou,E.(1999).‘Ananchoring
theoryoflightnessperception’.PsychologicalReview106(4):795–834.
Gilchrist,A.(thisvolume),InTheHandbookofPerceptualOrganization,editedbyJ.Wagemans.(Oxford:Oxford
UniversityPress).
Grossberg,S.,andMingolla,E.(1985).‘Neuraldynamicsofformperception:boundarycompletion,illusoryfigures,
andneoncolorspreading’.PsycholRev92(2):173–211.
Ho,Y.X.etal.(2008).‘Conjointmeasurementofglossandsurfacetexture’.PsycholSci19(2):196–204.
Hoffman,D.(2009).‘Theinterfacetheoryofperception:Naturalselectiondrivestrueperceptiontoswiftextinction’.
InObjectcategorization:Computerandhumanvisionperspectives,editedbyS.Dickinson,M.Tarr,A.Leonardis,
B.Schiele,pp.148–65.(Cambridge:CambridgeUniversityPress).
Kim,J.,andAnderson,B.L.(2010).‘Imagestatisticsandtheperceptionofsurfaceglossandlightness’.Journalof
Vision10(9):1–17.
Kim,J.,Marlow,P.,andAnderson,B.L.(2011).‘Theperceptionofglossdependsonhighlightcongruencewith
surfaceshading’.JournalofVision11(9),1–19.doi:10.1167/11.9.4.
Kingdom,F.A.(2011).‘Lightness,brightnessandtransparency:aquartercenturyofnewideas,captivating
demonstrationsandunrelentingcontroversy’.VisionRes51(7):652–73.doi:10.1016/j.visres.2010.09.012.
Page 14 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
Kingdom,F.,andMoulden,B.(1988).‘Bordereffectsonbrightness:areviewoffindings,modelsandissues’.Spat
Vis3(4):225–62.
Kingdom,F.,andMoulden,B.(1992).‘Amulti-channelapproachtobrightnesscoding’.VisionRes32(8):1565–82.
Lewontin,R.C.(1996).‘EvolutionasEngineering’.InIntegrativeApproachestoMolecularBiology,editedbyJ.
Colladoet.al.(Cambridge,MA:MITPress).
Marlow,P.,Kim,J.,andAnderson,B.L.(2011).‘Theroleofbrightnessandorientationcongruenceintheperception
ofsurfacegloss’.JournalofVision11(9):1–12.doi:10.1167/11.9.16
Marlow,P.J.,Kim,J.,andAnderson,B.L.(2012).‘Theperceptionandmisperceptionofspecularsurface
reflectance’.CurrBiol22(20):1909–13.doi:10.1016/j.cub.2012.08.009.
Metelli,F.(1970).‘Analgebraicdevelopmentofthetheoryofperceptualtransparency’.Ergonomic13:59–66.
Metelli,F.(1974a).‘Achromaticcolorconditionsintheperceptionoftransparency’.InPerception:Essaysinhonor
ofJ.J.Gibson,editedbyR.B.MacLeodandH.L.Pick,pp.95–116.(Ithaca,NY:CornellUniversityPress).
Metelli,F.(1974b).‘Theperceptionoftransparency’.ScientificAmerican230:90–8.
Metelli,F.(1985).‘Stimulationandperceptionoftransparency’.PsycholRes47(4):185–202.
Motoyoshi,I.,Nishida,S.,Sharan,L.,andAdelson,E.H.(2007).‘Imagestatisticsandtheperceptionofsurface
qualities’.Nature447(7141):206–9.doi:10.1038/nature05724.
Olkkonen,M.,andBrainard,D.H.(2010).‘Perceivedglossinessandlightnessunderreal-worldillumination’.Journal
ofVision10(9):5.doi:10.1167/10.9.5.
Olkkonen,M.,andBrainard,D.H.(2011).‘Jointeffectsofilluminationgeometryandobjectshapeintheperception
ofsurfacereflectance’.Iperception2(9):1014–34.doi:10.1068/i0480.
Paradiso,M.A.,andNakayama,K.(1991).‘Brightnessperceptionandfilling-in’.VisionRes31(7–8):1221–36.
Radonjić,A.,Allred,S.R.,Gilchrist,A.L.,andBrainard,D.H.(2011).‘Thedynamicrangeofhumanlightness
perception’.CurrBiol21(22):1931–6.doi:10.1016/j.cub.2011.10.013.
Rudd,M.E.,andArrington,K.F.(2001).‘Darknessfilling-in:aneuralmodelofdarknessinduction’.VisionRes
41(27):3649–62.
Shapiro,A.,andLu,Z.L.(2011).‘Relativebrightnessinnaturalimagescanbeaccountedforbyremovingblurry
content’.PsycholSci22(11):1452–9.doi:10.1177/0956797611417453.
Singh,M.,andAnderson,B.L.(2002).‘Towardaperceptualtheoryoftransparency’.PsychologicalReview
109(3):492–519.doi:10.1037//0033–295x.109.3.492.
Vladusich,T.(2013).‘Gamutrelativity:Anewcomputationalapproachtobrightnessandlightnessperception’.
JournalofVision13(1):1–21doi:10.1167/13.1.14.
Wallach,H.(1948)‘Brightnessconstancyandthenatureofachromaticcolors’.JournalofExperimental
Psychology38:310–24.
Notes:
(1)Thisreductionincontrastoccursforalmostanydefinitionofcontrast,whichincludesadivisivenormalization
termthatisafunctionofintegratedormeanluminanceintheregionoverwhichcontrastisdefined.Unfortunately,
thereiscurrentlynogeneraldefinitionofcontrastthatadequatelycapturesperceivedcontrastinarbitraryimages,
sotheprecisewayinwhichcontrastisreduceddependsonthedefinitionofcontrastusedinaparticularcontext.
Page 15 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015
The perceptual representation of transparency, lightness, and gloss
BartonL.Anderson
BartonL.Anderson,SchoolofPsychology,TheUniversityofSydney,Australia
Page 16 of 16
PRINTED FROM OXFORD HANDBOOKS ONLINE (www.oxfordhandbooks.com). (c) Oxford University Press, 2015. All Rights
Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a title in Oxford
Handbooks Online for personal use (for details see Privacy Policy).
Subscriber: OUP-Reference Gratis Access; date: 08 September 2015