PICTURE PERCEPTION: TOWARD A THEORETICAL MODEL · 2016. 11. 7. · J. J. Gibson's new theory of picture perception is described, and a program ... usually by differences in color,

Psychological Bulletin1974, Vol. 81, No. 8, 471-497

PICTURE PERCEPTION:TOWARD A THEORETICAL MODEL

MARGARET A. HAGEN *

Boston University

J. J. Gibson's new theory of picture perception is described, and a programof research within his framework is outlined. An analysis of pictorial informa-tion is proposed in which a systematic investigation of the structural com-ponents of pictures and their varying effects on perception is seen aspreliminary to the postulation of hypothetical pickup mechanisms. The basiccomponents of pictures are described, and literature is reviewed in the problemareas of distorted and impoverished information, observation from the wrongstation point, coexisting flatness and depth information, and the ambiguityof .the source of a single projection. The feasibility of the Gibsonian enterpriseis demonstrated, and further avenues for research into a structural analysisof pictorial information are pointed out.

A picture is a delimited surface withmarkings on it that represent something.This article is concerned with pictures in theWestern post-Renaissance mode; namely, thatattempt to represent, by means of structuralequivalence of some order, the layout of sur-faces in the world. Alternative modes ofrepresentation, perhaps requiring alternativeanalyses, are beyond the scope of this article.

By what means can a picture be said torepresent, to bring clearly before the mind,the segment of the world pictured? The aimof this article is to demonstrate the feasi-bility of extending J. J. Gibson's theory ofperception via motion-generated informationto an adequate theory of frozen pictorial in-formation and to explicate the problems thatmust be dealt with in any comprehensivetheory of the perception of pictures. Theproblems to be dealt with herein are the useby artists of modified linear perspective,caricature, impoverishment of information inoutline drawings and black-and-white photo-graphs, the coexistence of flatness and depthinformation, observation from the wrong sta-tion point, and the ambiguity of a singleprojection (i.e., the same projected form canarise from an infinite variety of shapes).

1 The author wishes to thank William M. Maceand Herbert L. Pick, Jr. for their conceptual assist-ance in the preparation of the original manuscript.

Requests for reprints should be sent to MargaretA. Hagen, Department of Psychology, Boston Uni-versity, 64 Cummington Street, Boston, Massachu-setts 02215.

THE GIBSONIAN MODEL OF INFORMATION

For Gibson the essential condition ofresemblance between pictures and the worldwas the correspondence of information con-tained in the structured light to the eye,coming either from a picture or from thereal scene. Gibson (1971) stated his positionquite succinctly in the following passage.

A picture is a surface so 'treated that a delimitedoptic array to a point of observation is made avail-able which contains the same kind of informationthat is found in the ambient optic arrays of anordinary environment [p. 31].

Central to Gibson's theory is the assertionthat picking up information about the per-sistent properties of objects and layouts ofsurfaces in the world is the grasping of in-variant structure. In order to consider theimplications of this view, it is necessary toexplicate more fully the ideas of structureand invariant information.

THEORY OF OPTICAL STRUCTUREAND INFORMATION

Gibson (1966) stated that ambient lightmust have structure if it is to carry informa-tion about (which means specification of) theenvironment. So whence comes the structureof the ambient light in an illuminated en-vironment? The ambient light reverberatingin the medium affords many possible stationpoints, or points of observation. The ambientlight converging on any one of these pointsbecomes the potential light to the eye. This

471

472 MARGARET A. HAGEN

light is structured, modeled if you will, bythe structure of the physical surface layoutof the environment and as such can conveyspecific information about it. The surfacelayout consists of the faces (with respect tothe observer) of surfaces and objects andthe smaller facets within them.

Boundaries between faces determine boun-daries in the light to the eye just as the formand slant with respect to the observer de-termine the angular projection of the light.Gibson argued that boundaries in the opticarray, in the light converging on a stationpoint, are due to the different amounts andcolors of light reflected from the surfaces ofobjects to that station point. Boundaries, orchanges in intensity, occur when two ad-jacent faces are at different angles of in-clination to the light source. Surfaces projectdifferent intensities if they have differentcomposition or pigmentation and thus produceborders in the optic array via their differentreflectance. These differences in reflectance,Gibson (1966) wrote, are further reinforcedusually by differences in color, doubly guar-anteeing borders in the light to the eye.

In short, an array is structured, i.e. caused to haveborders within it, by (1) the physical inclinationsof the faces and facets of surfaces, (2) the reflect-ance of the substances, and (3) the spectral reflect-ance of 'the substances, or chromatic reflectance, and(4) shadows [p. 194].

So slant, composition, color, and shadowdetermine borders in the light to the eye andtherefore specify borders of environmentalfaces and facets.

We can now assert a lawful relationshipbetween the light to the eye and the surfacelayout of the environment. The presence ofenvironmental information in reflected lightis defined, Gibson argued, by the univocalrelationship of a property of the stimulus toa property of the object. This is what hemeant rby the conveying of environmentalinformation. But determinate information(information specific to the persistent proper-ties of a particular layout) is invariant in-formation; that is, only through invariance(nonchange amidst change) can we arriveat specificity.

So far our postulated environment is sta-tionary and our observer, motionless. Gibson

(1966) pointed out that "the stationary per-spective of light does not specify the grossphysical layout of its source. The informationin the static array . . . is ambiguous in thisrespect [p. 199]." Movement of the observeris critical in resolving the ambiguities inher-ent in a static array because it provides op-portunity for the detection of invariant prop-erties of the array through the systematicperspective changes occasioned by motion.Motion perspective is the regular perspectivechange or optical flow of the texture elements(faces or facets) as the observer moves in theenvironment.

Slants, edges, corners, composition, andvarious separations of surfaces can be speci-fied in gradients of texture flow velocities orby specific types of discontinuities in theoptical flow. For example, Gibson, Olum, andRosenblatt (1955) have shown that the loca-tion, slant, and shape of surfaces as well asthe movement of the observer are specifiedin the structure of optical texture flow forthe case of straight line motion. Hay (1966)showed that the initial orientation and shapeof a flat surface is specified in the sequenceof projections of rigid motions of that surface.These distinctive characteristics of the opticalflow as the observer moves are specificallydetermined by particular characteristics ofthe physical environment and as such consti-tute information for that environment. It isonly through transformations of the opticarray, through motion of the observer, thatthe invariants in the optical flow specific toa particular surface layout can be detected.

PICTURES AS CARRIERS OF OPTICALINFORMATION

If the perception of the persistent prop-erties of objects and surface layouts is de-pendent on'the motion-generated informationpreviously described, then successful pictureperception seems to provide a serious embar-rassment to Gibson's theory of perception.Pictures as static, isolated, perspective in-stantiations of three-dimensional scenes can-not, apparently, act as occasions of suchmotion-generated information, so how then dothey function as carriers of information aboutsuch scenes? Evidence is subsequently re-viewed to suggest that pictorial perception is

PICTURE PERCEPTION 473

determinate not only with full-color, veridical,angle-for-angle, trompe-l'oeil representationsbut also with pictures suffering from theproblems of distorted or impoverished in-formation, observation from the wrong stationpoint, coexisting surface and depth informa-tion, and ambiguity of the source of the planeprojection.

How do we deal with this evidence in thebuilding of a coherent theory of picture per-ception? The field has gone in two directions.The more frequented avenue has been tobuild hypothetical structures of the mindcapable of perceiving the identity and locationof pictured objects through the addition orinterpretation of information gained throughpast experience. This avenue has been fol-lowed by Goodman (1968), with his theoryof pictorial structure as an arbitrary, con-ventional language that must be learned as askill, like reading; Hochberg (1968), who hashypothesized stored schematic maps evokedby feature pickup to account for picture iden-tification; Gregory (1970), with his theory ofevoked stored object hypotheses used inpicture recognition; and many others.

The alternative avenue, in the absence ofmotion-generated information, is to look notfor hypothetical mental structures but for thestructure in pictures; that is, for invariancein pictures, for what fails to change as theobserver moves or the slant of the picture ischanged. We can look for arrangements of,or relations between, lines and forms andtextures that remain invariant under trans-formation and hence provide the structuralbasis for determinate pictorial perception. Wecan look for the relations among picturedlines, forms, and textures that duplicate in-variant information in the real scenes pic-tured. Frozen three-dimensional informationis equivocal in part because its source maybe a picture and not a real scene; hence, suchreal world information may be captured ina picture (Pirenne, 1970; Smith, 1958).

Gibson (1971) has argued that picturesperform their representative function bymeans of just such an informational equiv-alence to the scenes they represent. Purdy(1960) and Gibson (1950) have suggestedthat one may look for pictorial structure inthe static rates of change (e.g., texture

gradients) in the absence of the rates ofchange of motion perspective information. Itis thus possible to separate descriptions ofpickup mechanisms from descriptions of whatis "pickupable," and indeed, given the stateof the field, it seems most advisable to do so.Until we arrive at a comprehensive, system-atic analysis and description of the structuresof different pictures and their varying effectson perception, hypothetical models of thestructure of the pickup mechanism seem pre-mature.

In the review of the different types of pic-torial materials to be discussed next, it ishoped that the feasibility of the Gibsonianenterprise is demonstrated and further ave-nues for research into a structural analysisof pictorial information pointed out. Sincethe data are meager and a systematic analysisof the problems of picture perception has notyet been undertaken, a precise empiricallybased model does not come out of this re-view, but the complex aspects of picturessubsequently discussed must, in the future,be embraced and resolved by any compre-hensive theory of the structure of picturesand pictorial perception.

BASIC PICTORIAL COMPONENTS IN WESTERNPOST-RENAISSANCE ART

According to Gibson, a picture succeeds asa representation of a real object or scenebecause it reflects the same structural in-formation in the light to the eye as the scenerepresented. Light carries information becauseit is structured, or caused to have borderswithin it, by the slant, composition, color,and shadows of surfaces in the world. Linearperspective is a necessary by-product of thisstructuring of light. It is a function of theproperties of light and the distance from thelight-reflecting object to the point of observa-tion. Light travels more or less in straightlines. As distance between object and ob-server increases, visual angle (size of retinalimage) decreases in a lawful gradient of pro-jected size. This diminishing projected sizeis true of all receding textured surfaces. Linearperspective is simply a special case of thistexture size gradient; it specifies the edgesof surfaces, the borders in the optic array.Texture gradients and therefore linear and


size perspective are information for distance,since they are determined by the distance be-tween observer and observed and the proper-ties of point projection. The consideration ofthe pupil and lens system does not invalidatethe information character of perspective.

Linear perspective has served as the pri-mary basis of depth information in Westernpost-Renaissance art. It is employed to rep-resent correct layout of the surfaces of ob-jects and correct spatial relations betweenobjects in depth. The success of linear per-spective as a source of information aboutthe depicted world rests, in part, on the as-sumption that the single perspective viewchosen must show the best or most charac-teristic aspect of the object or scene. Whensuch an aspect is not depicted, successfulrepresentation may depend on contextual orredundant information.

Linear perspective is not, however, theonly Western post-Renaissance discovery fortranslating the three-dimensional world to thetwo-dimensional picture plane. Relative sizehas also been employed as a depth cue in itsrole as a special case of a texture gradient.Superposition (overlapping or occlusion) isalso information for relative depth. Aerialperspective may be employed to give relativedepth via a gradient of clarity of outline andwarm-to-cold color gradations. Depth isfurther clarified and information for flatnessminimized by the addition of shadows. Shad-ows are used to indicate differential orienta-tion of surfaces to the source of illuminationand to mold the volume of a single surface(Arnheim, 1969; Gombrich, 1971; Wolfflin,19SO).

These traditional cues or formulae com-prise the model of representation that servedas the aim of Western art from the Renais-sance to the nineteenth century French rev-olution in painting. They still produce, forthe average Western adult, pictures that look"real" or lifelike. These formulae induce goodpictorial depth perception because as rules,as principles, they capture sources of depthinformation that are structurally similar tothe sources of depth information in the scenesrepresented. The classic cues of perspective(linear, size, texture, and aerial), superposi-tion, and shadowing do not represent simply

the limiting cases of motion-generated in-formation, nor are they arbitrary conven-tional symbols invented by artists. Rather,they may be considered in their own rightas informational components of pictures, bear-ing an obviously present (if unanalyzed)relation to three-dimensional reality, con-tributing to determinate picture perception,and as such requiring careful investigation.

The Western post-Renaissance model as-sumes the static situation. It assumes a frozenworld and frozen viewer at a single stationpoint (which must theoretically be the pointof projection for the picture). The model isalso burdened with a variety of viewing re-strictions made explicit by Gibson (1971)and discussed in the following section.

DISTORTED INFORMATIONModified Linear Perspective

Not everyone concerned with the makingof pictures agrees with Gibson's theory ofstructural equivalence as the essential aspectof the resemblance between the picture andthe depicted. Much of the argument aboutthe nature of the resemblance revolves aroundthe status of linear perspective. Goodman(1968) argued:

Representational customs, which govern realism, alsotend to generate resemblance. That a picture lookslike nature often means only that it looks the waynature is usually painted [p. 39].

He stated that pictures in perspective, likeany others, have to be read; and the abilityto read has to be acquired.

Gibson (1971) attacked the Goodmanlanguage view of pictorial art by attackingthe arbitrary conventionality of linear per-spective. He argued that the artist must, ofnecessity and not arbitrarily, use perspectivegeometry to transcribe the three-dimensionalworld onto a two-dimensional surface. How-ever, Goodman wrote:

In diametric contradiction to what Gibson says, theartist who wants to produce a spatial representa-tion that the present-day Western eye will acceptas faithful must defy the "laws of geometry" [1968,p. 16].

Goodman argued that not only are the con-ditions of observation of linear perspectivecumbersome and unnatural, but linear per-


spective itself, as traditionally used in art,is unfaithful to the true perspective of lightrays converging on a point. He stated thatartists and cameras systematically altercertain aspects of true perspective to producethe artistic linear perspective, which func-tions as a depth symbol in the Western modeof representation.

For Gibson the conflict resolved into aninadequate consideration on Goodman's partof the conditions of observation that mustaccompany the successful use of perspectivein art. According to Gibson the adequacyand validity of the laws of perspective havenot been challenged. For him the laws of per-spective are not conventions in any sense;they are a geometry, a property of light,and an optical and logical necessity. Whenthe laws of perspective are employed in pic-ture making, however, they are indeedsaddled with conventions.

Gibson formulated rules for observing thepicture surface: (a) It should be seen withone eye; (b) it should be upright and perpen-dicular to the line of sight, instead ofslanted, and its distance must be just suchthat the visual solid angle from the pictureis the same as would the visual solid anglebe from the thing pictured; and (c) thereshould be an aperture in front of the eyehiding everything but the picture itself. Henoted that these viewing prescriptions arealmost never followed in practice. He statedthat the system of perspective projection, theoptical geometry, must be distinguished fromthe practice of perspective. "Perspective dis-tortion" is produced by inadequate attentionto the conditions of observation and not bythe infidelity to the real world of the lawsof perspective. Since viewers could not bemade to abide by the prescribed conditionsof observation, the system was subsequentlymodified to reduce the resulting perspectivedistortion. The fidelity of the original systemof linear perspective, under proper viewingconditions, remains unchallenged.

The modified artistic perspective system,however, raises serious questions for thosewho would argue that there exists a non-arbitrary structural resemblance betweenpictures and the world, between linear per-spective portraying depth in a picture and

depth information in the real world. Can ob-servers tell the difference between modifiedand true perspective under the conventionalconditions of observation—that is, at theproper station point, with one eye, and witha window occluding the frame? Can they tellthe difference under normal conditions of ob-servation—that is, walking around in a gal-lery? Are there strict constraints on the de-gree of distortion required to produce an ap-parently faithful picture observed in a freesituation? It may well be that after an ex-haustive analysis of pictorial information theexplanation for picture perception may re-quire reference to some sort of pictorial modeof perceiving, learned through experience withpictures as representations of reality, as wellas to Gibson's theory of the structural-in-formational equivalence between the pictureand the depicted.

Caricature

Caricature has heretofore provided aspecial problem in picture perception be-cause of its nonconformity with point-to-point correspondence theories on the natureof the structural resemblance between thepicture and the depicted. A caricature makesno attempt to duplicate the light rays fromthe figure of the person represented, accord-ing to Renaissance principles of art, butsucceeds, nevertheless, in conveying informa-tion about the person portrayed. Gibson dealtwith the problem in his new definition of pic-tures as carriers of optical information aboutthe represented persons or scenes.

But the definition is broad enough also to admitthe case of a caricature, where the contrasts ofluminous energy are quite different, and even theforms are different, but where the high-order in-formation to specify a particular person is commonto both arrays. In short the optic array from apicture and the optic array from a world can pro-vide the same information without providing thesame stimulation. Hence an artist can capture theinformation about something without replicating itssensations J1971, p. 31].

Presumably the high-order information speci-fying an individual's face consists of suchrelations as the length or sharpness of theforehead curve relative to the length orsharpness of the nose curve, the width of theeyes relative to the length of the nose, and


so on. Kennedy (1971) quoted Gibson assaying that the information or invariants"are not to be found in the light rays andcolor patches, and not even in the forms ofthese elements, but rather in the forms ofform [p. 27]." That is to say, the structuralequivalence between a caricature and the per-son caricatured depends on the arrangementof forms and not on angular congruence orpoint-to-point correspondence. If perceptionconsists in part of the grasping of such in-formation as "forehead curve is very muchgreater than nose curve" or even "nostrils arevery wide relative to most other nostrils," andif it is this sort of relation information thatspecifies a particular face, then it follows thata drawing or caricature that preserves theserelations in portrayal will specify, even afterexaggeration, the same face.

There are certain constraints on caricature,however. For instance, if an individual'ssmall nose and wide mouth are more distinc-tive than his small nose and wide eyes, thena caricature exaggerating the former will bemore successful than one exaggerating thelatter. There may also be limits on how mucha relation can be exaggerated before it istreated as a new relation no longer specificto the subject of the caricature. The distinc-tive nose-mouth relation must be preservedunder the caricature transformation. Theproblem of which facial relation is mostdistinctive, most specific to a particular in-dividual, remains currently in the eyes ofthe caricaturist.

There is presently little or no research onthe subject of caricature. A careful develop-mental investigation of the means and limitsof caricature distortion would give us in-valuable information about facial perceptionand the sensitivity of the untutored to dis-tinctive features versus photographic detail.Shaw, Mclntyre, and Mace (in press) haveachieved promising success in the determina-tion of strain and shearing transformationsas specifying the: perceptual invariants ofboth aging and age levels; they intend toapply a similar analysis to caricature. Ryanand Schwartz (1956), in a very different ap-proach, investigated cartoons, the cousins ofcaricature. They ran an experiment to test

the effects of different media on the rapidpickup of pictorial information for such itemsas the posture of hands, the position of aswitch, etc. They compared a black-and-whitephotograph, an ink-and-wash-shaded drawing,a high-fidelity line drawing, and cartoons.The order of difficulty (easiest to hardest)was as follows: (a) cartoons, (b) photo-graph and shaded drawings, (c) line draw-ings. It may well be that the rapid pickupof information in a picture is best in picturescapturing the least amount of informationnot relevant to the specific task demands.Further research is needed, especially of adevelopmental nature. Gibson's theory maywell solve the problem of caricature but itcurrently does so mainly by assertion.

Impoverished Information

Outline drawings present a different prob-lem from caricature in that the informationthey present is impoverished, not distorted.Kennedy (1971) has defined outline figuresin his thesis.

A tentative definition of these is that they are line-figures formed by continuous lines with no areasof "shading," i.e. there are no large areas, usuallymarked off by lines, whose width is several timeslarger than the lines in the drawing, which havebeen covered or "filled in" with pigment. In outlinefigures the length, direction and shape of every lineis significant [p. 2].

Gibson (1951) made the point that outlineforms geometrically represent the margins ofa solid form or the edges of a surface formand that they have two margins instead ofthe single margin given by the edge of anobject. Line drawings preserve only the infor-mation for the edges of objects, for boun-daries in the optic array. These boun-daries show discontinuities of optic layoutdue to changes of color or slant or texture ofsurfaces or the presence of shadows. Thesesources determine contrasts in the opticarray, but any contrast can be due to anyone of these sources. Kennedy noted,

So any particular contrast, considered on its own,is ambiguous with respect to its environmentalorigins. However, there may be information in theoptic array, in terms of the arrangement of contrastsrather than the isolated contrast, to specify theorigin of any particular contrast [1971, p. 31].


Kennedy, however, stated that we do not yetknow enough to discuss in any detail thestructure or arrangement of contrasts. Weknow they exist. We don't know how to de-scribe them and we don't know how theyoperate in perception. Their role in resolv-ing pictorial ambiguity in outline drawingsmust be crucial.

If contrasts in the optic array are de-termined by changes in slant, reflectance,composition, or shadowing, how is it thatthe change itself can be specified by a lineon paper independent of any portrayal ofthe surfaces, reflectances, or shadows de-termining the change, except by means ofthat same line on paper and its position orfunction in an arrangement of lines? Ken-nedy simply remarked, "It can be said thatlines have two contours, and so give rise totwo contrasts in projected optic arrays[1971, p. 32]." Two contrasts specify twochanges, that is, at least three sources ofdifferential intensity in the optic array..It doesn't necessarily follow from thepresence of areas of change in the opticarray that a line drawn on paper willgive rise to a perception of that change.However, granting the possibility, Kennedy'sargument continues along the following lines.Because a contrast in the optic array maybe projected by any number of sources ofchange in surface layout or shadowing, sotoo, perhaps, may a line in a line drawingrepresent any one of those features. Since acontrast in the optic array presumably givesrise to a determinate perception of changethrough its position in an array or arrange-ment of contrasts, it is feasible to undertakeresearch on the types of information thatcan be picked up in line drawings. It shouldbe noted, however, that the sources of de-terminate perception in line drawings neednot be the same sources as those that giverise to determinate perception in a colored,textured picture or a three-dimensional scene.

Kennedy then, with Gibson, was arguingthat edge, or boundary, information is funda-mental to perception of surface layout. If apicture captures that edge information, thenit should serve as a source of information forsurface layout. The perception of surfaces inthe world, however? is, multiply determined

both by motion-generated information andthe co-occurrence of sources of contrast witha single surface change. For example, as theeye moves from observing one object to ob-serving another, it is unlikely in the naturalworld that the change of surface will occurwithout an accompanying change in color ortexture or slant. Nonoccurrence of suchsimultaneous change (e.g., a change in slantwithout accompanying changes in texture andcolor) indicates the several surfaces of asingle object. Outline drawings lack thiscrucial redundancy of information about sur-face change, but Kennedy argued that theresolution of the contrast ambiguity lies inthe arrangements of the lines specifying thosecontrasts. He suggested a program of researchinvestigating responses to arrangements oflines to achieve a theory of that resolution.Kennedy's own thesis question along suchlines was, "Can line-segments in line-draw-ings represent basic features of surface lay-out [p. 4] ?" His answer was mostly in termsof demonstrations for the adult reader, sim-ilar to Gibson's (1969a) observations in"Three Kinds of Equivocal Information inLine Drawings." Such demonstrations do notprovide answers to the basic questions of thenature of the structural relation between out-line drawings and their subject matter. Isboundary information in a line drawing asufficient source of information for the un-tutored observer, a very young child, or amember of a culture with no outline draw-ings? Must the relation between outline draw-ings and the world they picture be learned;must the relation between outline drawingsand other pictures be learned? Do they reallypreserve enough information for unambiguousperception by the naive observer? Much ofthe developmental research on picture per-ception makes use of the impoverished in-formation in outline drawings or black-and-white photographs to test the ability of theuntutored to recognize pictures of familiarobjects, to isolate or complete the contourof single objects, and to see depth or thelayout of surfaces in pictures. In the briefreview that follows, the impoverishment ofinformation in black-and-white photographsis clarified.


Recognition of Pictured Objects

For the simple recognition of pictured ob-jects, data from infants are far from conclu-sive. There is, however, one study of a 19-month-old child explicitly on recognition offamiliar objects in pictures. Hochberg andBrooks (1962) reared their child until the ageof 19 months with extremely restricted ex-posure to pictures and no exposure to picture-plus-naming experiences. At 19 months he wasable to successfully identify simple and com-plex line drawings and photographs of famil-iar objects. The line drawings always (excepton one presentation) preceded the photo-graphs of the same object in the task. Ofoutline drawings they wrote:

"Ghost shapes," as Gibson has called them, may beanemic, but they are by no means deceased . . . thecomplete absence of instruction in -the present case(the absence of "association" between picture andrepresented object) points to some irreducible mini-mum of native ability for pictorial recognition. Ifit is true also that there are cultures in which thisability is absent, such deficiency will require specialexplanation; we cannot assert that it is simply amatter of having not yet learned the "language ofpictures [p. 628]."

That some African natives have not yetlearned the "language of pictures" was theassumption made by Segall, Campbell, andHerskovits (1966). They quoted Hersko-vits:

I have had an experience of this kind, similar tothat reported from many parts of the world bythose who have had occasion to show photographsto persons who had never seen a photograph before.To those of us accustomed to the idiom of therealism of the photographic lens, the degree ofconventionalization that inheres in even the clearest,most accurate photograph is something of a shock.For, in truth, even the clearest photograph is aconvention; a translation of a three-dimensionalsubject into two dimensions, with color transmutedinto shades of black and white. In the instance towhich I refer, a Bush Negro woman turned a photo-graph of her own son this way and that, in attempt-ing to make sense out of the shadings of grays onthe piece of paper she held. It was' only when thedetails of the photograph were pointed out to herthat she was able to perceive the subject [p. 32].

Whether or not such "data" are any good,Segall et al. have a point about the impover-ishment and conventionality of black-and-white photographs. They are certainly two-

dimensional representations of real scenes andas such are sources of both types of informa-tion. They are less impoverished than outlinedrawings, however, because they preservetexture and shading information for thespecification of sources of contrast in theoptic array. Representation of color changesin the optic array as shades of gray is ofcourse a convention, although there certainlyexists a predictable relationship between theshades of gray and the color changes. Such arelationship may have to be learned, althoughHochberg and Brooks (1962) indicated thatit does not.

Segall et al. offered an anecdote on pictureperception in relatively pictureless environ-ments.

It is interesting that the experience of anthro-pologists shows that motion pictures are almostuniversally perceived without trouble and thatcolored prints are also—although here the naiveteof the respondents may be questioned in more recentfield experience [1966, p. 33].

However, evidence contradictory to thisstatement was found by Nadel (1937). Hecompared the picture perception of the Nupe,a people with imageless art, with the Yoruba,a people with art rich in images, using black-and-white photographs and found no failuresof picture detection in either group, althoughidentifications were frequently rather idiosyn-cratic or culture bound.

Deregowski (1968) cited Brimble (1963),who submitted men and women from Barotse-land to an identification test consisting of twosheets of "simple line drawings," 58 in all.Correct identification responses ranged from94.3% to 98.7%. Deregowski pointed outthat it is impossible to tell if the rare errorswere due to "detection" errors (failure torecognize the representative nature of thedrawing) or "identification" errors (namingthe wrong object).

In Deregowski's own work (1968) in ruralZambia, he found that both school childrenand men were above chance on recognitionof photographs of models of familiar animals,but only children were above chance on rec-ognition of pictures of unfamiliar animals.His data suggest that errors were in identifica-tion or inability to identify and not due to


detection failure. Similarly, Mundy-Castle(1966) in Ghana and Kilbride and Robbing(1968) in Uganda found many "mi&-identifications" of outline drawings of an-imals and a road but no failures to perceivethe outline drawings as representations ofthree-dimensional objects and surfaces.

As a last note on "primitive" art, the workof Dennis (1960) deserves mention. Hestudied the human figure drawings of illiter-ate Bedouins living in the Syrian desert.They have no native forms of representa-tional art and only minimal exposure toWestern art forms. They could draw picturesof men both in outline and filled in butusually in a rather sticklike form. Althoughthere is no clear relation between people'sdrawings and their perceptions of drawings,it is interesting that a people with almost noexperience with representative art of any sortcould accept and produce a variety of "con-ventions" in drawing human forms. Dennisstressed the impoverishment of their draw-ings, the lack of facial features, clothing, andother detail, but this very impoverishment,this schematization, is intriguing. Thereseems to be a spontaneous acceptance of theevocative power of lines and outlines, of therepresentation of a large object on a smallpiece of paper, of a round object on a flatsurface. Such acceptance does not fit wellwith Segall et al.'s assertion of African in-ability to recognize the representative char-acter of black-and-white photographs. Ofcourse, the Bedouins may be an unusualgroup, but their relative innocence of repre-sentation was well established. More workneeds to be done as Segall et al. (1966)noted:

Here is an area in which systematic research shouldbe done to support anecdote, for naive respondentson whom 'to try such experiments may now or willsoon be lacking [p. 35-36].

Davenport and Rogers (1971) have goneeven further than the African work in theirinvestigation of the ability of the untutoredto recognize black-and-white photographs ofobjects. Two chimpanzees and an orangutanhad previously been trained to match a visualsample with a haptic object. In this studythey were required to match a photograph

with a haptic object. They had no familiaritywith photographs. Correct choices with colorphotographs ranged from &0%-WO% across40 different pictures. With black-and-whitephotographs, choices were 60%-100% cor-rect. "We can now conclude that apes canperceive a photograph of an object at firstsight [p. 320]." If this finding is replicable,it certainly suggests a fundamentally evoca-tive character for photographs, despite theirmultiple dissimilarities to reality, and arguesstrongly against a theory of pictures as anarbitrary language whose rules must belearned through experience.

Overlapped, Embedded, and FragmentedContour

Bower (1966) used infants 50-60 days oldwho had never seen a triangle before exceptas they occur in the real world. He condi-tioned them to a wire triangle crossed withan iron bar. Infants generalized most to aplain uninterrupted triangle. He repeated theexperiment with slides and there was no pref-erence for one test figure over any other.

As in the case of space perception, the.infants' per-formance appeared to depend not on static retinalcues but rather on the information contained invariables, such as motion parallax, that are avail-able only to a mobile organism viewing a three-dimensional array [p. 90].

Interpreting baby studies is always diffi-cult, but Bower's data seem to suggest thatmotion parallax information is necessary todetermine the identity of an object across thesuperposition transformation. Motion parallaxappears to be needed to establish the identityor unity of the triangle alone when crossedwith a bar. For very young infants, over-lapped pictured objects seem to be seen asflat, not layered in depth. Bower's data sug-gest that it is possible to ignore or fail toperceive depth information in a picture whileattending to the information for pictures asflat objects (single surfaces).

Ghent (1956) studied the performance ofchildren 4-13 years old on overlapping andembedded figures. She found that young chil-dren exhibit little difficulty in analyzingoverlapping figures, regardless of whetherthey are realistic or geometric, and rather

480 MARGARET A, HAGEN

great difficulty in analyzing embedded figures.She suggested that

when a figure was so "hidden" by other forms thatthe boundaries of the added forms coincided withthose of the original figure, the figure would beharder to find than when the boundaries of theadded forms intersected with those of the originalfigure... it could be said that the improvementwith age reflects an increase in the capacity to per-ceive a boundary as belonging to more than onefigure [p. 587].

Since her figures were outline drawingswith no background, it is extremely difficultto generalize her findings to other types ofpictures, more complete representations.Coincidence and intersection of boundaries,however, certainly occur in the real worldand in the world pictured. Her findings withrespect to embedded figures are an interest-ing comment on Kennedy's (1971) assertionthat lines have two contours and hence giverise to two contrasts in the projected opticarray. The younger children's difficulty inanalyzing embedded figures suggests thatthey see only one contour and thus one con-trast. It may be that contrast ambiguity inline drawings is a learned perception or atheoretical figment and that the arrangementof lines in line drawings is far more deter-mining of perception than previously sup-posed.

Bower's (1966) data suggest that a pic-tured object whose contour is partially over-lapped by another object is not completed,or filled in, by infants. Ghent's (1956) data,on the other hand, show that children asyoung as four years have little trouble per-ceiving pictured objects with overlappingcontours. This does not, however, reflect adevelopmental change in the ability to fill inor elaborate impoverished information be-cause in Ghent's case contour was intersectedbut not disrupted, and the surface of theobject was not occluded by another depictedsurface as it was in Bower's study. Ghent'soverlapping study was a study of sensitivityto transparency rather than to superpositionor occlusion information.

There seems to exist no programmatic re-search on the development of sensitivity tothe multiple forms of pictorial information(linear perspective, size perspective, super-

position, texture gradients, aerial perspective,shadows and highlights, color contrast, trans-parency, caricature, and arrangements offorms) be it with high-fidelity, colored, tex-tured pictorial stimuli or impoverished out-line drawings and black-and-white photo-graphs. We have, rather, patches for a crazyquilt insufficient for piecing together. Twomore such patches on contour fragmentationare mentioned before moving to a brief re-view of studies on perception of pictorialsurface layout with impoverished information,

Gollin (1960) used line drawings to testfor age changes in the amount of informa-tion needed in identification of outline draw-ings. His "subjects were nursery school andkindergarten children. The younger childrenrequired more completion than the older onesfor successful recognition, but even their per-formance was extremely good considering thefragmentation (the contour interruption) ofthe outlines. Even very young children seemto assume continuity where there is .none.A careful structural manipulation of com-ponents can determine under what conditionssuch assumptions occur, what possible con-figurations of line segments give rise to theunification of the segments into a perceptualwhole. The urge to do so must be quite primi-tive, perhaps because chaos and meaningless-ness are unnatural or unperceivable.

Mooney (1957) studied closure ability inchildren aged 7-13 years using some ex-tremely interesting stimuli. The stimuli wereblack-and-white drawings producing the effectof strongly lighted high-contrast photographsof heads and faces showing only salient shad-ows and highlights. He found an increasingability with age to recognize and classify thefaces - by age and sex. This may reflect animproved ability to recognize higher orderrelations of shadows and highlights specify-ing faces. A more complete analysis of thestimuli and the configurations that facilitaterecognition would be most desirable.

Surface Layout and Depth

There have been very few investigations ofsurface layout and depth in outline picturesand black-and-white photographs, and whatwork has been done is mainly of one type—studies qf African pictorial depth perception,


Hudson (1960, 1962a, 1962b, 1967) was theprime mover of a whole minor body of re-search. He devised the Hudson pictorialdepth perception test, which has since beenused repeatedly by himself and other in-vestigators. The test consists primarily of out-line drawings of three figures: a hunter, anantelope, and an elephant. Depth was de-picted variously by relative size, overlap, andlinear perspective.

Responses to the questions whether the hunter wasaiming at elephant or antelope or whether elephantor antelope was nearer the hunter were taken asself-evident indications of two-dimensional or three-dimensional pictorial perception [1967, p. 94].

The consistent result seems to have been thatsubjects did best whose cultural backgroundand home life favored frequent exposure topictures. Hudson thought the two-dimensionalperformance was due to misperceptions of the"conventional pictorial depth cues." It is im-possible, however, to tell if nondepth percep-tion was due to the inability to perceive depthinformation, the greater salience of the in-formation for flatness, the general poorquality of the drawings, or the specific termsused in questioning subjects, as suggested byOmari and Cook (1972). A wider sampling ofstimuli and testing conditions is needed tocheck the validity of the task as an index ofthe ability to see depth in outline drawings.

Two other points should be noted. Giventhe simplicity of the test, it is odd that nosample, white or black, young or old, pro-duced anything near perfect performance.This result indicates that it may not be thesubjects or the pictorial depth cues that areat fault but the depth portrayal in Hudson'spictures. Also, with black high school pupilsand graduate teachers Hudson (1960) stated,"Hesitation in responding was noticeable andwas particularly pronounced with the gradu-ate teachers, some of whom took as long asone hour per picture to respond [p. 203]."One hour per picture to respond to two simplequestions suggests that these black samplesdid not view the task in quite the samemanner as Hudson did. Better communicationbetween the experimenter and the subjectswould have given the results greater credi-bility.

Deregowski (1968) repeated Hudson's testwith Bantu adults and children in Lusaka.He also gave them line drawings representingwire figures to be constructed by the sub-jects. Even where subjects did not make three-dimensional responses to Hudson's pictures,they frequently constructed three-dimensionalwire figures from the line drawings. Makingthe wire figures first improved performanceon Hudson's pictures. These apparently con-tradictory results cast some doubt on thevalidity of Hudson's test as a measure ofpictorial depth perception.

Mundy-Castle (1966) used the Hudsontest with Ghanaian children 5-10 years old.Nearly all the subjects almost always gavetwo-dimensional responses. There were alsomany misidentifications of objects like theroad, the horizon, the hill, and the elephant.Now clearly if the elephant is seen as a "pig,rabbit, goat, sheep, lion, tiger, dog" or othersmall animal, then all the resultant size-depth relations in the picture are altered.Depth in these pictures is not very deter-minate even with correct identification; with-out it the test is useless as a depth perceptionmeasure.

Another point, even granting perfect com-munication, perfectly drawn pictures, anderrorless identification of objects, is that al-though one can draw in perspective, one neednot. In Hudson's pictures, elephant, man, andantelope are all drawn in direct side viewsexcept that one cannot tell the front of thehunter's body from the back. Two converg-ing lines are then inserted to give these cut-outs relative depth. The point is not musttwo lines representing perspective alwaysevoke depth, but can they? When will thesubject see such lines as perspective lines, notdoes he always and must he regardless ofother elements in the picture, to be consid-ered a three-dimensional picture perceiver?

A pictured scene is a unit. Piecemeal ma-nipulation of its parts will almost invariablyresult in odd distortions of the overall rela-tions. A certain amount of impoverishmentof information is required before such ma-nipulation is even possible. (Witness the lackof texture and shading in Hudson's pictures.)Something must determine that the picturedroad is a road, that the horizon line does


represent the horizon before depth relationsbecome determinate. The Hudson test failsto analyze the conditions for determinatepicture relations. Six depictions of three depthcues simply cannot analyze which com-ponents or combinations under what condi-tions will act as sources of depth information.Dismissing the cues as culture-bound con-ventions does little to further understanding.

All of the investigators using Hudson'stests seem to deny the basic ambiguity of thepictures. For instance, Kilbride and Robbins(1968) used two of Hudson's pictures to testBagandans' (Uganda) ability to use linearperspective as a pictorial depth cue. Eachsubject was asked for each road, "What isthis?" A correct response (e.g., road, river,path, etc.) was interpreted as use of the linearperspective cue to pictorial depth perception.An incorrect response (e.g., hill, stone, lad-der, letter "A", etc.) was interpreted as afailure to use the linear perspective cue fordepth. This was not a test of Bagandans'ability to use linear perspective as a depthcue. It did indicate their inability to acceptunequivocally two lines drawn on a card asthe edges of a road. "Correct" response wasthe experimenter's response, not intrinsic un-ambiguous truth.

As a concluding note, Hudson stated thatschool-attending white children acquire adultresponses to his pictorial depth perceptiontest gradually between the ages of 6 and 12years. The credibility of this statement isquestionable as previously noted, althoughthe developmental literature, with one ex-ception (Stern, 1924), does indeed suggest agradual development of the ability to pickup pictorial depth information. Stern (1924)found that his daughter Hilde, one and a halfyears old, would imitate actions representedin pictures and could interpret real size fromthe relative sizes of pictured objects. Sheand other less-than-two-year-olds could alsorecognize objects drawn in perspective.Bower's work, on the other hand, showed aninsensitivity in infants to pictorial informa-tion for size-at-a-distance.

Bower (1965) conditioned infants 40-60days old to respond to a white paper cube.Another group was conditioned to respond

to a slide projection of that cube equal tothe real cube in visual angle, luminance, andso on. There were three subsequent gener-alization conditions: a change in size, achange in distance, and a change in size anddistance such that the retinal image was equalto that of the training stimulus. Infantstrained on the real cube discriminated allthree conditions from the training stimulus(or at least responded differentially). In-fants trained on the slide responded differ-ently to the change in size and to the changein distance than they did to the trainingstimulus, but they failed to discriminate thelarge far cube, projecting a retinal imageequal to the training stimulus, from the train-ing stimulus. That is, they responded tomeasurable area on the projection screenand not to portrayed size. Pictorial cuespresent but insufficient for making the dis-crimination were: (a) relative height on theprojection screen, (b) density of table tex-ture at point of cube contact, (c) density ofwall texture occluded by the cubes, and (d)linear perspective of the table to point ofcube contact. Bower concluded that motionparallax is the basic source of size-in-depthinformation for babies and that sensitivity topictorial cues comes in later. This researchdoes not tell us that an infant cannot recog-nize a picture of its mother, bottle, or otherfamiliar object, but it does suggest that amother pictured near is not the same as amother pictured far, that cube pictured nearis not seen as identical to the same cubepictured far.

Wohlwill (1962) tested children in gradesone, four> and eight and adults on line draw-ing perspective displays. He presented themwith rectangles at different pictorially de-fined depths. The children adjusted the"near" rectangle larger than the "far" rec-tangle to achieve an apparent size match andvice versa. Even the youngest children weresensitive to perspective depth information.Wilcox and Teghtsoonian (1971) tested chil-dren three and nine years old and adults.They placed equal area figures at differentpictorially defined depths on slides to pro-duce the illusion of a "large" object "farback," that is, high up on the picture plane.They used everything from outline drawings


like the Ponzo illusion to magazine illustra-tions. Three-year-olds showed no respond-ing above chance to the illusory larger ob-ject; nine-year-olds showed 13% respondingand adults, 90%. This suggests that in-sensitivity to pictorial cues such as foundby Bower in babies may persist beyond theage of three and still have some effect onadult performance. Wilcox and Teghtsoonianfailed, however, to test for the effects ofvaried motion parallax information, and theirdata may show not the ineffectiveness ofpictorial size cues but the effectiveness ofmotion parallax as a cue to flatness. Theyalso failed to vary and control for the rela-tive effectiveness of the cues (texture densityand regularity, superimposition, and rela-tions among various cues or components)employed in the drawings and photographs.Also, the geometric pairs may have beenperceived as floating in space rather thanresting on a surface because they were flatfigures without three-dimensional form or at-tached shadows. The failure of adults to re-spond 100% correctly to pictorially definedsize in depth may also suggest that therewas something wrong with the slides. Becausethe figures were pasted on the slides, theymay have looked in some way "unnatural."

Summary

The studies of recognition of pictured ob-jects in outline drawings and black-and-whitephotographs by the untutored overwhelm-ingly suggest an unlearned responsiveness toimpoverished representations. This supportsGibson's and Kennedy's arguments on thefundamental importance of edge informationfor perception of surface layout in the naturalworld and in pictures. The evocative natureof line drawings and black-and-white photo-graphs of objects has been demonstrated,but we still lack a systematic attack on thesources of the evocations, on the arrange-ments of lines and contrasts that give riseto determinate perceptions of specific repre-sentations.

For overlapped, embedded, and fragmentedfigures the picture is more confusing. Infantsdo not seem to complete interrupted contourwith surface occlusion. Nursery school chil-dren are quite good at completing interrupted

contour without surface occlusion. Four-year-olds successfully perceive figures throughoverlapping transparencies (intersected, non-interrupted contour without surface occlu-sion), and between the ages of 7 and 13 thereis an increasing ability to employ shadowand highlight information specifying faces.The need for programmed research is obvious.The preceding research gives only the baresthints about the evocative nature of variousline arrangements and the development ofresponsiveness to such arrangements.

Research on perception of surface layoutand depth with outline drawings and black-and-white photographs similarly gives littleground for a firm theoretical stand on theevocative character of the line. The Africanwork, in general, suggests a gradual develop-ment of the ability to see depth in suchpictures, but the test and testing conditionsas previously discussed are too questionablefor either theory testing or building. TheWestern data suggest that sensitivity to linearand size perspective information is absent ininfants, present in children less than two, ab-sent in three-year-olds, and thereafter presentbut improving to 90% sensitivity in adults.Such conclusions are hardly warranted, how-ever, given the state of the field, and as longas we lack a comprehensive task analysis andprogrammatic developmental research theanswers will not be forthcoming.

Gibson treated the developmental prob-lems of picture perception within the frame-work of naive versus pictorial attitudes inperception (1969b, 1971). Gibson stated thatthere is evidence to suggest that young chil-dren, animals, and prepictorial men do notnotice the appearance of an object (an aspector perspective at a single stationary pointof view) or the perspectives of the environ-ment. Children do not notice the appearanceof the environment as a frozen patchworkof flat colors confined by the boundaries ofthe temporary field of view. "Instead, theynotice only the crude distinguishing featuresof objects that are given by invariants oftransformation in time [1969b, p. 9]." Bythe information carried in the light from thestructured world, children see whole objects,not aspects or appearances of objects, andtheir relation to one another in the surface


layout of the visual world. This is the naiveattitude.

The pictorial attitude, the ability to see theappearances of a frozen world, perspective ona two-dimensional plane (not depth in athree-dimensional world) is a later develop-ment for children and for man. When manbegan to draw his world he had to learn tosee its two-dimensional appearance; likewise,when a child begins to draw he must learnto draw, learn how to see his world pic-torially as a flat bounded surface.

If all of the above is true, as Gibson wouldhave us believe, what are the developmentalimplications of such a theory? If it is as-sumed for the moment that the child mustlearn to see appearance, to see the world asflat and stationary, in order to draw or paintthat world and that adults can switch be-tween the naive and perspective attitudesonce taught the latter, what does this sayto us about children's ability to pick up in-formation in pictures for surface layout beforethey have learned to draw? What man mustdo to create pictures may say very littleabout what he must do to perceive them.Gibson's theory is not, however, a model forinformation pickup. His chief endeavor hasbeen to describe what is "pickupable," nothow it is done. That is, Gibson was con-cerned with describing the structure of thelight to the eye, but the extension of histheory into models of the pickup mechanisminvolves consideration of the structure of thehead. Because Gibson himself has not takenthat step, the problems arising from it donot lie at his door. Nevertheless, there aresome problems and they deserve brief men-tion because they must eventually be con-sidered by any coherent theory of picture per-ception.

The theory of structural equivalence be-tween the picture and the depicted is quitepromising (although preliminary) in its at-tempt to describe and account for the basisof adult picture perception but is less suc-cessful in describing the development of suchperception in the child. Once the invariantshave been detected, once the distinctive fea-tures of real objects have been discriminated,then the basis of pictorial perception has been

established; but how does the child acquiresensitivity to such information or features?How many times and from how many differ-ent perspectives must a child see a particu-lar cube, not to mention how many differentcubes, before he can recognize any member ofthe family of cubes from any perspective ina picture? What are the constraints on theability to do so and do such constraintschange with age? Are distinctive features ofclasses of objects hierarchically salient? Isthere a developmental change in the positionof features in such a hierarchy? Are depthcues hierarchically informative, that is, canthey be ranked according to the strengthwith which they determine the perception ofdepth relations in a picture? Are their effectsin combination different from their effectswhen depicted singly? Are redundancy andcontextual determination diminishing needswith experience? How do these two factorsinteract with the necessity of representingan object or scene in its best or most char-acteristic aspect? Is it true that any repre-sentation can be disambiguated with sufficientpictorial context, perception-determining ar-rangements of lines and forms?

We may also need to posit some sort ofnested hierarchy of invariant relations. Thatis, there must exist "knowledge" at somelevel of how the invariant relations specify-ing cube differ from those specifying rec-tangle, sphere, and so on. When do suchdiscriminations and their interrelationshipsbecome clear to children? In order to arguethat the untutored perceive representative artsuccessfully because of its structural equiv-alence to reality, it is necessary to showsome basis for assuming their sensitivity tothe prerequisite structural information in theworld. We need carefully controlled develop-mental studies of sensitivity to structuralfeatures of the natural world via motion-generated information, sensitivity to thosesame real world features in the static case,and sensitivity to that static informationisolated and manipulated in pictures. Somepreliminary work along these lines is nowreviewed.


The Coexistence of Surface and

Depth Information

The problems of distorted and impover-ished information are of a particular order.They deal with the content of a picture andits relation to the world. They attempt toanswer the question of how the observerknows what is being represented via thecontent of the vehicle of representation. Atthis level the central issue is the equivalenceof the structural information in the light tothe eye coming from the world and fromwithin the picture. But the problem of theways and means of representation presup-poses a prior level of the act of representation.To say that a picture is a framed piece ofcanvas or similar surface with paint orsimilar markings on it that represent some-thing presupposes the knowledge, on somelevel, that A can represent B, that a picturecan represent the world, regardless of themeans of representation. When and how doesthe child come to know that object A is aparticular object representing another objectB (or an arrangement of objects, a segmentof the world)? When is a picture accepted asanything other than a framed surface witha flat array of colors and forms? It is clearthat at this point the two levels of the rep-resentative act, (a) knowledge that A canrepresent B and (b) A represents B bymeans of similar information, become soentwined that the explication of one is de-pendent on the explication of the other, atleast for pictorial representation.

Perception of the visual world occurs bymeans of information. If pictorial representa-tion occurs by means of that same informa-tion, in what sense must knowledge of thepossibility of pictorial representation occurprior to the successful perception of the con-tent of pictures? In other words, if the twosources of information (the world and pic-tures) are identical, then no act of perceptionof representation is necessary, but only an-other act of perception of the real world. Inthis sense, a knowledge of the possibility ofrepresentation need not be postulated priorto the perception of pictorial content becauseno such knowledge is necessary for perceptionof the real world, and the two types of per-

ception occur by means of the same informa-tion. There is, however, one major difficultywith this dismissal of the problem of repre-sentation. The formulation only holds for theperfect picture observed under perfect con-ditions of observation. We can modify thedemand for perfection a little, as Smithshowed with his photomural of a corridor in1958, and the result is almost the same. Ob-servers perceive not the picture but the pic-tured world. No knowledge of the possibilityof representation is required because thepresence of representation cannot be detected.In such a case, from the point of view of theobserver, presentation, not representation,occurs. Thus we need posit no prior knowledgeof the possibility of representation. Gibson(1951) stated,When it is carefully arranged that the picture isseen through an aperture so that the frame is in-visible, the head is motionless, and only one eyeis used, the resulting perception may lose Us rep-resentational character [p. 406].

Difficulty arises, however, because this case,the perfect picture under perfect conditionsof observation, is the nonexistent case exceptin experimental laboratories. Thus, its use-fulness is somewhat limited in explainingother cases of picture perception. The per-fect case does, however, highlight the in-tertwining of the two levels of representation,possibility and means. Under the perfect case,the means obviated the necessity of postulat-ing prior knowledge of possibility.

In less-than-perfect cases, the less-than-perfect picture viewed under less-than-perfectconditions of observation, the means of rep-resentation become far more complex thanstraightforward duplication of real world in-formation. As previously noted, many com-ponents of pictures bear an obvious struc-tural relationship to the world they repre-sent—for instance, color, texture gradients,linear perspective, aerial perspective, super-position, relative size, and arrangements oflight and shadows, not to mention the littleunderstood relationships among the severalcomponents of a single optic array. Our ob-server, however, is now holding the picturein his hands, turning it over, and viewing itwith two eyes from the wrong station point.Real world information must in some sense


be duplicated in this picture, but there area good many other sources of informationspecifying the picture only as a rather flat,perhaps framed, surface with splotches ofcolor on it. As Gibson (1951) stated, "Pic-tures stand for substantial objects in additionto' being substantial objects [p. 409]."

How pictures stand for substantial objects,how they carry information for the realworld, has already been discussed exten-sively. Information for a picture being a sub-stantial object comes from many sources.Motion parallax tells the observer the pictureis flat thus: For a picture held upright inthe frontal-parallel plane, the entire picturetranslates at the outside edge of the frameacross the extrapictorial background as aunit, and there are no differential rates oftranslation of objects across the backgroundwithin the picture (no gradients or discon-tinuities of optical flow). Stereopsis and in-formation for the surface of the plane ofprojection (e.g., the texture of the canvas)also tell the observer the picture is only atwo-dimensional surface with a pattern ofvarying luminosity or pigmentation. The tex-ture of the canvas, projection screen, and thelike, the mismatch of lighting between slidesand the surround, and the mismatch of thepicture surface with the surrounding surfacesall are sources of information for the flatnessof the picture.

Herein lies the central problem of recog-nizing that a vehicle of representation ispossible. Because both information for thepicture as a rather flat object like any otherobject and information for the pictured scenecoexist in a picture, can one or the other beignored or unperceived? If the latter is per-ceived, then perception of the pictured con-tent takes place; if the former is perceivedas well, then perception of a picture as anobject that can act as a vehicle of representa-tion also takes place. Picture perception isperceiving that "this is a picture of a lakewith swans up close and sailboats far away."It is not mistaking the picture for a reallake or seeing only a framed blue surfacewith white patches on it. If this is true, thenpicture perception demands response to bothsources of information, and such a responseentails knowledge of the possibility of repre-

sentation even if that knowledge first occursonly upon the first presentation of a pictureto the naive observer.

Granting, then, the presence of coexistingand conflicting flatness and depth informa-tion in pictures, the question of point ofobservation, or station point, must be raisedto rescue the observer from the horns of thisdilemma. Pirenne (1970) treated this ques-tion explicitly. He observed that most ordin-ary pictures are almost invariably viewedfrom the wrong station point, that is, a pointof view different from the correct center ofprojection of the picture or photograph.Nevertheless, observers successfully perceivethe pictured objects and relations of objects.In such cases, the linear perspective containedin the picture is no longer duplicating thelight coming to the eye from a similar realscene. If perspective remains a valid effec-tive source of information even when theobserver is at the wrong station point, itmust do so because of hitherto unexplainedhigher order relations between the views fromthe right and wrong station points. Besidesthe theoretical deformations that should re-sult from improperly observed linear perspec-tive, Pirenne also noted that at a stationpoint sufficiently different from the correctone, the picture flattens. The scene depictedno longer appears in three dimensions. Hisexample, however, was a painted churchceiling at a considerable distance from thespectator. With an ordinary picture the sta-tion point disparity needed to flatten thescene would necessarily be much smaller.,Kennedy (1971) argued that given a knowl-edge of the correct station point, the trans-formation of the optic array, its differencefrom the standard, could be mathematicallydetermined, although the mathematics wouldnot be trivial.

The implications for perception for an ob-server at the wrong station point are far fromclear, yet it is probably true that the wrongstation point is the most frequently occupiedone. Can such perception really be unlearned?Can it be said to follow naturally from someas yet undetermined relation between theright station point and all of the wrong ones?Can the one-eyed observer even find the rightstation point and if not, why not? Does an


observer notice the deformations and flatten-ing noted by Pirenne (1970)? If the validityof linear perspective is explained in terms ofthe properties of light and the laws ofprojection, does this explanation hold whena fundamental axiom of the picture projec-tion system is violated (i.e., the observer notat the correct center of projection)?

Smith (1958), using a photomural of acorridor, found that adult subjects couldestimate the number of paces from their view-point to a specific part of the pictured sceneand the number of paces between differentplaces in the scene. The scene was viewedat two distances, two feet and nine and onehalf feet from the eye of the subject, monocu-larly.

When the photograph was viewed at a distance of9-i feet, the corridor appeared to be over twice aslong on the average as when the photograph wasviewed from a distance of 2 feet [p. 81].

Smith did not discuss the relation of thesedistances to the correct station point, butthe magnification-ramification effect suggeststhey were on the normal from the correctstation point to the picture surface. Smith'sstudy was the beginning of an answer to thequestion of what happens to pictured rela^tions when observed from the wrong stationpoint.

Smith and Gruber (19S8) attempted amore systematic analysis of the magnification-minification effect of changing the stationpoint on the normal previously denned. Theyfound that when the photomural was viewedfrom the correct station point, impressionsof depth were approximately the same forthe photograph and the real corridor. Atother viewing distances apparent depth wasreduced by approximately one half whenthe image was magnified two times. That is,as the viewing distance increased, the imagegrew smaller and the perceived distancegreater, in a geometrical relation.

Farber (1972) investigated the effect ofangular magnification on sequences of rigidoptical motions, and demonstrated that suchmagnification may produce consequent non-rigid sequences. The simple magnification ofinformation alters the nature of that informa-tion, in this case.

We still need information, however, on theeffects of wrong station points other thanthose on the normal from the correct stationpoint to the picture plane. Given a sufficientlydisparate station point, pictured forms shouldappear distorted and the scene depicted shouldflatten. Yet when most spectators view mostordinary pictures, neither of these eventsoccurs. Pirenne (1970) suggested an "intuitiveprocess of psychological compensation" toaccount for this. "[Ordinary] pictures pro-duce an illusion of a very particular kind,to which we become accustomed as part ofour education [p. 162]."

Leaving aside such exceptional cases as Pozzo'sceiling, peep-holes, and ordinary trompe-l'oeilpaintings (in which invariably the subject matterhas very little extension in depth) ordinary picturesviewed binocularly in the usual manner do not givea genuine 'three-dimensional representation of theirsubject matter . . . . In spite of the fact that it maygive a very strong suggestion of depth, the represen-tation given by ordinary pictures is of a sui generisnature, and depends on the spectator's subsidiaryawareness of the characteristics of the picture surface[p. 166].

The intuitive process of psychological com-pensation "is based both on the spectator'sawareness of the surface of the picture, andon his preconceived ideas regarding the com-ponents of the scene presented [p. 162]."What Pirenne meant by preconceived ideasabout the picture's components is unclear,but Polanyi clarified the notion of surfaceawareness in his introduction to Pirenne'sbook. Polanyi made a strong distinction be-tween trompe-1'oeil paintings and ordinarypaintings. In tromp-1'oeil art there exists nosubsidiary awareness of the canvas, of thesurface of the painting, no knowledge of rep-resentation enters in. The distortions andflattening are easy to observe from the wrongstation point. There is little flexibility in theconditions of observation. Because a trompe-1'oeil painting is regarded not as a painting,not as a representation, but as the "realthing," distortions are striking when obtainedviews violate expectations. With ordinarypaintings, there is no such illusion of reality.

The trompe-1'oeil artist succeeded in producing acompelling illusion of a three-dimensional scene;whereas, however effective the perspective of anormal painting may be, the picture will not be


mistaken for the sight of its real objects. Its ap-pearance has depth and this is an essential featureof it, but this depth is not deceptive: it is felt tohave flatness in it. It has the quality of a mixtureof depth and flatness [p.. xvii].

The Pirenne-Polanyi thesis is that ordinarypictures provide a very special kind of illu-sion. They provide information for both flat-ness and depth, an unusual duality to whichwe become accustomed through education andexperience. Successful perception of the depthrelations in a picture is dependent on a sub-sidiary awareness of the picture's surface,on the flatness information, and on the per-ception of a representative event.

Developmental Utilization of CoexistingSurface and Depth Information

This definition of picture perception, how-ever, may well describe only an end pointand bypass developmental steps on the way.The flatness and the depth information in apicture may be considered to jointly de-termine the perception of a picture as anobject that represents another object or scene,as suggested by Pirenne (1970) and Polanyi(1970), but they may also be thought of aspotentially in conflict during the developmentof picture perception. E. J. Gibson (1969)suggested that the discrepancy between thepictured scene information (in depth) andthe information for flatness given by theabsence of this motion parallax (gradientsand discontinuities of flow velocities) in pic-tures is ignored by the sophisticated viewerwho attends to the former for the perceptionof the pictured scene and ignores the latterexcept to perceive a picture not a replica.

All of the previously cited studies failedto take into account the coexistence in apicture of flatness (or surface) and depthinformation. Yonas and Hagen (1973) in-vestigated children's sensitivity to pictorialdepth in a replication and extension of thework of Wilcox and Teghtsoonian (1971). Anattempt was made to test E. J. Gibson'scontention that the flatness information in apicture must be minimized for successful per-ception of portrayed depth by the young oruntutored. Following E. J. Gibson's sug-gestions, Yonas and Hagen compared theperformance of three-year-olds, seven-year-

olds, and adults on judgments of relative sizein photographic transparencies, tinder twodifferent conditions of observation. Half ofthe subjects viewed the slides in a conditionof potentially conflicting information. Theslide edges were visible and the subject's headwas freely moving; thus, motion parallaxinformation for the flatness of the slide wasavailable. The other half of the subjectsviewed the slides under restricted conditionsof observation designed to minimize conflictby decreasing the information for flatnessavailable. The slides were viewed througha peephole, thus eliminating head motion,and through a window placed between thepeephole and the screen, so that it cut offview of the slide edges. Thus, in this condi-tion there was no motion parallax informa-tion for flatness.

Adults and three-year-olds showed no con-sistent difference between the two conditions.Only seven-year-olds showed a consistentlyinferior performance in the conflict conditionwith free head movement. It may be that ourmanipulations failed to sufficiently diminishthe conflict. It is true that even in the non-conflict condition, the slides looked like slides,not real objects in depth. Our results withthe nonconflict slides should have paralleledthose obtained in the control condition em-ploying real objects viewed through a peep-hole. They did not. At all ages and withall visual angle relations pictured, perform-ance on the slides was worse than performanceon the real object control. There seemed tohave been sufficient information for flatnesseven without motion parallax. It should benoted, however, that although performanceon the slides did not depend on the presenceor absence of motion parallax information,it did depend very strictly on the visualangle relations portrayed. As the visual anglesize relations of the two objects picturedwent from equal to the reverse of the absolutesize relations, subjects increasingly respondedto the visual angle relations rather than tothe real size relations. .This indicates an in-creasing tendency to react as "if the slidesportrayed only a flat surface without depth.Subjects increasingly responded to the amountof area on the screen occupied by the photo-graphed object. Still, no age group's perform-


ance ever reached a perfect error rate. Sub-jects always seemed to respond to the realsize differences as well as to the visual anglerelations, perhaps vacillating from trial totrial. As the visual angle relations becamemore drastically the reverse of the real sizerelations, performance became more and moreunder the control of the former.

Apparently then in size judgments in slidestwo things are going on. One is the effect ofthe visual angle relations portrayed, and theother is the conflict between depth informa-tion and flatness information. The two effectscan be factored out by comparing perform-ance on slides with that obtained in real ob-ject controls, with and without head motion.Our results suggest that it is possible to pro-duce a continuum of responding increasinglycontrolled by the visual angle relations pre-sented rather than the real size of the por-trayed objects. This result may not obtainwith binocularity.

A better setup is needed to control for anddiminish the amount of information availablefor flatness in slide viewing. E. J. Gibson(1969) might be right in arguing that onecan induce good pictorial depth perceptionin young children by minimizing flatness in-formation, and we simply failed to reducethe conflict enough to test her hypothesis.On the other hand, Pirenne (1970) might bsright. Back-projected slides on a glass screenmay offer few surface cues while still provid-ing both flatness and depth information. Atthe right station point, an observer may notrequire an intuitive process of psychologicalcompensation for distortions, but he may stillneed to be in a pictorial or representativemode of perception. If it is necessary to ac-quire knowledge of the peculiar kind of illu-sion presented by pictures, then it must benecessary to know when to apply that knowl-edge. Knowing when may depend on aware-ness of the picture's surface. That is, aperception of flatness may trigger a com-pensation for flatness. Back-projected slidesmay provide an insufficient source for thisrequisite information, leaving observers tornbetween the two types of information pre-sented. In Yonas and Hagen (1973), thesurface information present was ambiguous.There was too much information to parallel

a trompe-l'oeil or a real scene and too littleinformation to trigger consistent perceptionin a pictorial mode. The surface of a slidelooks both insubstantial and present.

The amount of information for surfaceavailable may be scaled from least to mostin three situations: (a) trompe-l'oeil artparalleling real scenes, (b) slides viewed underordinary conditions, and (c) photographicprints. E. J. Gibson (1969) and Pirenne(1970) agreed that picture perception is alearned ability. Gibson noted that the sophis-ticated viewer attends to the information fordepth and ignores the flatness information.Thus, for adults in this culture she shouldpredict no difference in performance in situa-tions varying in amount of surface informa-tion available. The performance of youngchildren would presumably depend on theirlevel of sophistication with pictures. At veryyoung ages, performance should depend onthe amount of discrepant information avail-able: Performance with trompe-1'oeil artshould be better than that with ordinaryslides, which in turn should be better thanperformance with photographic prints.

In Yonas and Hagen (1973), however,adult errors reached 25% with certain typesof stimuli and 10% with others. Adults wereconflicted. They were clearly not ignoring allthe flatness information or their slide per-formance would have been error free, as itwas in the real object control condition. E. J.Gibson (1969) overestimated the abilities ofthe sophisticated adult, at least when thepictures in question are slides. She might beright, however, about the effectiveness ofminimizing flatness cues to enhance depthinformation. Successfully carried out, suchmanipulations would produce a trompe-l'oeilsituation that would be conflict free. It isquestionable, however, whether partially suc-cessful manipulations would perform a sim-ilar function to a lesser degree.

Pirenne (1970) would probably argue thatbetween trompe-l'oeil art and prints or paint-ings exists a conflicted limbo into which slideswould surely fall. Surface awareness is pres-ent but in a ghostly form. Spectators respondneither as to trompe-l'oeil nor as to an or-dinary painting, both of which courses wouldlead to perfect size-in-depth relation respond-


ing. Rather, observers vacillate between theflatness information and the depth informa-tion. Their vulnerability to flatness cues inthis conflicted situation presumably de-creases as their experience with pictures in-creases. Thus, children should respond to theflatness information to a greater degree thanthe adults, as indeed they do. Even adult per-formance, however, should be conflicted sincethere is no unambiguous information presentin a slide to tell them how to respond. Afterall, given two present but conflicting sourcesof information, something must determinewhich receives a response.

An investigation was designed to test theimportance of awareness of the pictorial sur-face and of position of station point in ac-cordance with the Pirenne-Polanyi theoryof artistic perception. As previously argued,photographic transparencies provide a spe-cial kind of pictorial depth ambiguity. Theprojection surface is neither completely • im-perceptible, as in trompe-l'oeil setups, nor isit obviously perceptible, as it is with photo-graphic prints or paintings normally viewed.The representative quality of slides may beambiguous. Pirenne and Polanyi would pre-dict that adult performance improves withan obvious picture surface.

Since a fairly regular scaling of surfaceinformation from trompe-l'oeil to ordinaryslides to photographic prints was desirable,an initial study was designed to test theeffectiveness of a trompe-l'oeil slide-viewingapparatus. It was postulated that true trompe-l'oeil responding duplicates responding to areal frozen scene viewed through a peephole.Hence, the study was designed to see whichcondition in Yonas and Hagen (1973) pre-dicted the results in the proposed trompe-l'oeilapparatus. In Yonas and Hagen adults madeno errors when viewing real scenes through apeephole. When viewing slides through apeephole, they made 3%, 12.5%, and 27%errors on the equal visual angle, 80%visual angle, and 70% visual angle slides,respectively.

This study was not a replication of Yonasand Hagen since the stimuli were colored andthe background was regular, not random. Thestimuli were also more numerous and a

greater variety of sizes and distances wereused. It was felt, however, that the condi-tions were sufficiently similar to warrant com-parison. In the trompe-Poeil study, adultsmade 3.9% errors on the equal visual angleslides, 28.9% errors on the 85% reversedvisual angle slides, and 39.8% errors on the70% reversed visual angle pairs. The resultsdo not in any way suggest the error-free orlow-error rate responding expected in atrompe-l'oeil condition. The control by visualangle relations apparent across all ages withslide viewing in Yonas and Hagen is evidentlyoperative in the trompe-l'oeil study. Accord-ingly, the attempt to devise a trompe-l'oeilapparatus was abandoned for the present, andthe attempt to scale surface informationavailable was confined in the next studyto a comparison between ordinary back-projected slides and photographic prints.

The task was similar to that used in Yonasand Hagen. Subjects were asked to choosethe larger of two objects of similar shapephotographed against a textured background.Squares were paired with squares and tri-angles with triangles. The objects were po-sitioned against the background at distancesyielding three types of visual angle relationsbetween members of a pair: (a) pairs inwhich the visual angles of the two objects(large and small) were equal, (b) pairs inwhich the larger object had a visual angleonly 85% that of the really smaller object,and (c) pairs in which the visual angle ofthe larger object was only 70% of the visualangle of the really smaller object. There were16 slides and photographs of each kind ofpairing. The subjects were four-year-olds,seven-year-olds, and adults.

The study was designed to test the develop-ment of sensitivity to the point of observa-tion and to look for developmental evidenceof a pictorial mode of viewing triggered bystation point and surface information. If per-spective is a valid cue to distance, to whatextent is its effectiveness constrained by theartistic rules for observation? If linear per-spective can be successfully used irrespectiveof the correct station point, then its effective-ness must be dependent on the perceptionof higher order invariants, such as gradients,independent of their direct projection to the


OSui0a.

100

75

O 50

u&u.O 25

36

LU

gPseudo-Trompe-l'oeil

• Normal Picture Viewing"'"S Triggers Pictorial Moae

Abnormal

Right Wrong Right Wrong Right Wrong

STATION POINTADULTS 7-YEAR-OLDS 4-YEAR-OLDS

FIGURE 1. The interaction of age with surface information and station point.

eye. This perception must be learned sinceperspective views of the real world mustnecessarily be projected to the eye and to noother point. Only in art does a visual scenehave a center of projection other than theeye. To test station point sensitivity, the testdisplays were viewed both from the correctstation point and from an incorrect stationpoint forty degrees to the left of the correctstation and at the same distance from thescreen.

The Age X Surface X Station Point inter-action reached significance only at a fairlylow level (0.05 < p < 0.10), but its trendsare highly suggestive of hypotheses for futurework (see Figure 1). Adults seem to treatphotographic transparencies and photographicprints in very different manners. Slidesviewed from the correct station point, al-though not identical to trompe-l'oeil presenta-tion, nevertheless induce the best pictorialdepth responding in adults. The situationseems to provide enough trompe-l'oeil in-formation so that no pictorial mode need betriggered nor flattenings compensated for.This is not, however, the normal experience ofviewing pictures. Pictures, both slides andprints, are normally viewed from the wrongstation point. When a picture is observedfrom the wrong station point, the observer

may be tripped into a pictorial mode ofviewing and the compensation for incorrectstation point misprojections triggered. Thus,the data suggest that the adult viewers treatboth slides and prints the same way whenobserving from the wrong (the normal) sta-tion point. The data also suggest that thesize-in-depth perception of adult viewers innormal pictorial situations (wrong stationpoint) is inferior to that obtaining withpseudo-trompe-l'oeil representations (rightstation point, minimal surface information).Likewise, prints viewed from the correctstation point provide the adult observer withan abnormal situation. Size-in-depth responsesare near chance since he is consistently inneither the pictorial mode nor the trompe-l'oeil (real world) mode of perceiving, andhis performance reflects his indecision.

Overall, four-year-olds responded to pic-tured depth only about a quarter of the timeand their pattern of responding was verydifferent from that shown by the adults.Four-year-olds responded to the pictorialsize-in-depth in slides and prints at the cor-rect station point and in slides at the incor-rect station point all at approximately thesame low level. Although station point didnot affect size-in-depth performance withtransparencies, depth responding was congid-


erably diminished when prints were observedfrom the incorrect station point. Four-year-olds, unlike the adults, cannot use the stationpoint cue and surface information to switcha pictorial mode of viewing and thus areincreasingly hampered as more and moreflatness information is added by introductionof observation from the wrong station pointand of information for the texture of the papersurfaces of photographic prints.

The pattern of results for the seven-year-olds lies between that of the adults and thethree-year-olds. There is a suggestion of adevelopmental change in susceptibility toboth flatness of surface information and shiftin station point. As with the adults, slidesat the correct station point induce in seven-year-olds the greatest number of correct size-in-depth responses, and the seven-year-olds'responses to depth in photographic trans-parencies is diminished by observation fromthe incorrect center of projection. Slides mayoffer insufficient surface information to trig-ger a compensation for the flattening of thepicture observed from the wrong stationpoint. With prints, however, the data suggestthe beginning of a pictorial mode of viewingtriggered by observation from the wrong, butmost frequently occupied, station point.Seven-year-olds may be beginning to treatprints observed from the right station pointas the abnormal situation, neither trompe-1'oeil nor in the normal pictorial mode.

Under observation from the wrong stationpoint, projective distortions do occur and arereadily observable in photographs of photo-graphs. Why then are they not perceptiblein ordinary viewing? Pirenne (1970) hasargued that perception of the surface andcomponents of the picture triggers a com-pensation for the perspective distortions. Thepresent study concentrated on manipulationof available surface information and stationpoint to test Pirenne's hypothesis, but thedata are only suggestive of such a compensa-tion mechanism. At least two explanationsother than the compensation hypothesis mayexplain successful picture perception fromthe wrong station point. First, the observermay fail to notice the pictorial distortionssimply because he cannot do so without theinterjection of a pictorial medium; that is,

pictures of pictures taken from the wrongstation point and subsequently observed fromthe wrong station point. Perhaps there issimply enough elasticity in the visual systemso that observers are unable to notice a sta-tion point switch. The trends in the datamay simply reflect randomly different be-haviors in different conditions across differ-ent ages. Perhaps, too, size judgments maybe rather impervious to such a minor ma-nipulation as a 40-degree station pointchange. Shape judgments or matching mayprovide a more sensitive index of the effectof station point through the theoreticallypresent perspective distortions. If the systemis sufficiently elastic, observers should beunable to notice even the shape distortionsacross a fairly wide range of station points.

Second, whereas the degree of elasticityjust suggested may not be present in thevisual system, it is certainly not unfeasible toargue for the redundancy of information inthe components of the picture. It may wellbe that the perception of a variety of surfacerelations is so overdetermined that a mis-perception is unlikely even at the wrong sta-tion point. We don't yet know enough todescribe the correspondences and codeter-minations of the relations in a single opticarray, but it seems highly probable that theyaffect perception in a way unavailable tosingle cues in isolation.

Summary

The Pirenne-Polanyi hypothesis of a pic-torial compensation mechanism has been sug-gested to account for successful picture per-ception under the normal conditions of ob-servation from the wrong station point andmay, by extension, account for successfulperception of modified linear perspective.Goodman's (1968) argument for the neces-sity of modifying linear perspective to achievecorrect perception of pictured relations, how-ever, is somewhat inconsonant with the sug-gestion of such a learned compensation mech-anism. Who compensates? Is it artist, ob-server, or both? The Gibson-Kennedy argu-ment for the power of "forms of forms," fordeterminagy in the arrangement of the pic-torial components, may well dispense withthe need, for either postulated compensation.


for successful pictorial perception. Until theproblems of form distortion and depth flat-tening have been systematically attacked bythe pitting of one such theory or postulateagainst another, no resolution of the ques-tions is possible. Gibson's "forms of forms"sounds, on the surface, simplistic because wepresently lack a careful analysis of redun-dancy and codetermination in the frozen opticarray. Yet such an approach may well re-lieve us of the need to hypothesize compen-sation mechanisms and bring the problem ofpictorial perception back to an analysis ofthe structure of the light coming to the eyefrom the picture. Even Pirenne made briefnote of the role of the observer's ideas aboutthe components of the picture in overcomingdistortions and Battenings. Pirenne's "pre-conceived ideas" about pictorial componentsmay perhaps be better translated into Gib-son's notion of the perception of the severalcodetermining aspects of the frozen opticarray rather than into the syntax of Good-man's learned picture language. The arrange-ment of forms in an optic array (in a pic-ture) may also ultimately resolve the dilemmaof the fundamental ambiguity of any singleprojection, to be considered next.

Multiple Generation of Frozen Information

All pictures are fundamentally ambiguousbecause each is a single frozen perspective ofan infinitely large possible set of visual scenes.In pictures there is no disambiguating mo-tion parallax, no walking around to the otherside to see. Because there is no possibilityof seeing the other sides of objects in pic-tures, the identification of pictured objects isalways necessarily ambiguous. The back of apictured cube may well be concave, in whichcase it is no longer a cube. This seems atrivial ambiguity, particularly because it isalso encountered in the real world, albeitwith fewer constraints on checking the per-ception, but it is not. Successful perceptionof the form of an object from a single frozenperspective implies that perception can occuron the basis of a "best guess" made withonly a subset of the features necessary todistinguish the true form of an object. Acareful structural analysis of co-occurringfeatures of objects and a contextual analysis

of surface layout would help us account forsuch perceptual weights as assumptions ofregularity. It would be helpful to know whichaspects of a single perspective view of ashape, both in and out of context, give riseto assumptions about unseen surfaces.

Recall that any contrast in the optic arraycan be due to any combination of differentialfacing, reflectance, color, and shadowing. Thesource of a single contrast is ambiguous.That is, a cast shadow may look like a changein color, a cast shadow may look like a changeof facing, a change in color may look like achange in slant, and so on. In the real world,gradients of optical flow velocities, occasionedby the texture of a surface as the observermoves, provide information sufficient tomake the extension and slant of a surfacedeterminate. Pictures, however, provide nosuch source of additional information sincethe motion perspective occasioned by the ob-server's movements is irrelevant to the per-ception of pictured surface relations. Indeed,as the observer moves in front of a picture,the single gradient of texture flow velocitiestells him that the relation the pictured sur-faces have to one another is one of coplanaradjacencies, a piece of information most un-helpful to the perception of pictured surfacelayout. How then can an observer know thesources of contrasts in a pictured optic ar-ray? As previously suggested, to resolve theproblem of impoverished information in out-line drawings and black-and-white photo-graphs, we must (even with fully colored,textured pictures) fall back on such difficult-to-grasp, but no less valid, variables as ar-rangements of contrasts and forms of forms,or, in other words, the components of thepicture.

The lack of motion parallax informationin pictures also gives rise to theoreticallyambiguous slant and relative depth. Kennedy(1971), in his exposition of Gibson, haspointed out that most surfaces are homogen-eously textured, and a discontinuity of thetextured light indicates a discontinuity ofenvironmental texture, a change of surface.

Hence, adjacent solid angles of optic texture in theoptic array, that lie inside discontinuities of optictexture, will have come from adjacent surfaceareas of the environment [p. 8].


Thus, we know that a change in texture indi-cates a change in surface, but we do not knowfrom the configuration of the solid anglealone whether the surface slants or remainsat a constant distance from the station point.How can we tell a slanted cube from atruncated pyramid?

While a particular surface area in the environmentwill project a particular solid-angle element to agiven station point, the same configuration of thatsolid angle could be projected by different surfacesin the environment [Kennedy, 1971, p. 8].

Kennedy appealed to the gradient of atextured surface occasioned by recession ofthat surface from the station point to dis-ambiguate slant. If a surface ds homogene-ously textured, then the 'farther a textureelement is from the eye, the smaller will beits projection by the light to the eye. Thetexture of the surface as a whole will bepacked more densely, and its borders willapproach each other, in accord with a lawfulgradient, as the surface slants away from thestation point (Purdy, 1958). Goodman(1968) has taken issue with the determinacyof this information in an attack on artisticlinear perspective. He pointed out that thesame texture gradient can be observed anddrawn when a surface rises in space or goesdown in a hole, as well as when a surfacerecedes along the z axis. A texture gradientin a picture tells us the surface is at a slantto its plane of projection, but nothing in thetexture gradient alone can tell the observerthe slant with respect to the ground planeor with respect to the observer himself. Evenif the observer views the picture from itscenter of projection, slant remains ambiguouswithout an assumption of congruence betweenhis orientation and the picture's orientation.

A pictured surface is not just slanted withrespect to the observer, resulting in the per-ception of "this is a surface at a slant withrespect to me." Pictorial slants and arrange-ments of slanted surfaces must be specifiedas well by the information giving the locationof the ground plane within the picture. Tex-ture gradients alone cannot give this sort ofuseful information about surface arrange-ments in pictures without recourse to othercomponents of the picture or assumptions

carried by the observer. For example, Ben-son and Yonas (1973) found sensitivity inthree-year-olds and adults to shadow andhighlight information for concavity and con-vexity. The three-year-olds, however, showedthis sensitivity only when the" pictures wereshown vertically, upright in the frontal-parallel plane. Adults consistently assumedthe direction of illumination to be at the"top" of the picture, be it horizontally fartheror vertically higher, whereas the three-year-olds assumed a source of illumination in keep-ing with a gravitational reference system;that is, from the sun or light above. Theadult behavior is an excellent example of alearned assumption held by the observer. Onthe other hand, in the same study three-year-olds showed great sensitivity to linear per-spective information for size-at-a-distance inGibson's (1950) cylinder drawing, even whenthe picture was rotated 180 degrees, render-ing the surface relations depicted most un-likely.

In the face of the difficulties with texturedpictorial surfaces, the next point is rathertrivial and has previously been discussedquite extensively. Whatever problems of am-biguity textured drawings suffer, the prob-lems occasioned by the further impoverish-ment of outline drawings are far more num-erous. That human beings so readily acceptdrawn lines as representing clear borders inthe optic array is no less than astonishing.As Gibson (1951) remarked,

an outline, representing as it does only the edges ofa surface, may stand for any object which projectsthat particular outline, including some very queerlyshaped surfaces [p. 410].

In the same article, Gibson reported thatsubjects' responses to simple outline figuresalmost invariably fell into one of two classes:(a) objects for which the outline form wasa plan view, and (b) objects for which theoutline form was a perspective view. Anytheory of picture perception would benefitgreatly from a theory of the pictorial com-ponents determining which type of percep-tion occurs. If it is possible to set up anambiguous situation, then it should be pos-sible to disambiguate it in an eventually law-ful manner. The type of ambiguity referred


to here should more properly be called equiv-ocality because the information is not strictlyambiguous but- rather calls forth more thanone perception.

CONCLUDING REMARKS

The position expressed in this article isbasically in agreement with Gibson's newformulation of the order of structural re-semblance between pictures and the worldthey depict.

As reasonable as Gibson's new definitionof pictures seems, problems with it comeimmediately to mind. What is informationin the world like? Is it static and frozen sothat its replication in pictures is fairlysimple, or is it dynamic, dependent on move-ment of the organism through its environ-ment? It is possible to make a case for eitherposition.

This article assumed, with Gibson, thatmotion-generated information is fundamentalto the perception of surfaces and their layoutin the world. If motion-carried information isfundamental, however, where does that leavethe position on pictures as environmentalinformation duplicators? One can talk of thestatic case as the limiting case for motion in-formation, but it is not particularly helpful.One can also construct elaborate models ofhypothesized mental structures to account forthe pickup of frozen information. This articlesuggests that a more fruitful pathway forresearch is the systematic investigation ofpictorial information in the Gibsonian termsof ecological optics. It is hoped that thefeasibility of this enterprise has been demon-strated. The information analysis approachdoes not dispense with the problems of dis-torted or impoverished information, coexist-ing surface and depth information, or pictorialambiguity, but it does suggest a specific lineof investigation of their effects on percep-tion. A pictorial mode of perceiving patternedafter Pirenne's (1970) and Polanyi's (1970)suggestions was briefly considered, but it ishoped that any such hypothetical compensa-tion mechanisms will remain only in thebackgrounds of theories of picture- percep-tion, until the data from an exhaustive in-vestigation of the information components

of pictures demand the use of such con-structs.

There seems to be insufficient informationin pictures to result in unambiguous identi-fication of pictured objects or perception ofpictorial surface layouts. We know, however,that a 19-month-old child can recognizefamiliar objects in outline drawings andphotographs without previous training andwith very restricted exposure to pictures. Weknow that untutored Africans and youngchimpanzees can do the same. We have onlyfragmentary information on the ability ofthe untutored to perceive pictures despiteoverlapped, embedded, and fragmented con-tour.

We know very little about the perceptionof surface layout and depth in pictures. TheAfrican research suggests that it depends onhow much one's cultural background andhome life favor frequent exposure to pictures.American studies indicated that adults arevery good at estimating distances and other-wise responding to the depth relations por-trayed. The African data is not such as onewould wish to build a theory on, and theAmerican studies have not gone far enough,especially in their failure to systematicallystudy the development of pictorial perceptionin children. We know very little about adultperformance on pictures with an eye to dis-covering which theoretically present ambig-uities and problems influence responding andwhich fail to do so. We know almost nothingabout children's growing responsiveness topictures or about the development of theability to recognize and use representations.

This article argued that picture perceptionis perceiving "this is a picture of a lake withswans up close and sailboats far away," notmistaking the picture for the real thing orseeing only a framed blue-and-white surface.That is, in order for a picture to be perceivedas such, not only must it carry informationduplicating the information in the real scene,but it must also carry information for itselfas a picture. This moves the problem alonga different avenue. We no longer want toknow if pictures can fool people into thinkingthey see the real world, but how do picturescome to act as representations of the realworld and by what means do they do so?


In order to successfully perceive pictures,the naive observer must accomplish severaltasks, (a) He must learn to disregard theunnatural stillness of the frozen world pre-sented to him by a picture and make thebest use of the information remaining, (b)He must learn to make the best guesses aboutsize, shape, and distance in the face of im-poverished information or ambiguity, per-haps based on past experience with the worldor, as Gibson would have us believe, on thecodeterminance of the several components ofan optic array (the "forms of forms") pro-viding both specialized and redundant in-formation, (c) The naive observer must learnto understand and to use certain conventionswhose relationship to the world they repre-sent is symbolic, without the physical cor-respondence of two similar structures (in theworld and in the picture) modeling the lightto the eye in a similar manner, (d) He mustlearn to disregard the station point with itsattendant deformations and flattening, (e)He must accommodate himself to the coex-istance of flatness and depth information ina picture either through suppression of theflatness information or through its use, asPirenne suggested, as an aid in resolving theambiguities, distortions, and flattenings of afrozen array observed from an incorrect sta-tion point.

REFERENCES

Arnheim, R. Visual thinking. Berkeley: Universityof California Press, 1969.

Benson, C., & Yonas A. Development of sensitivityto static pictorial depth information. Perceptionand Psychophysics, 1973, 13, 361-366.

Bower, T. G. R. Stimulus variables determiningspace perception in infants. Science, 196S, 149,88-89.

Bower, T. G. R. The visual world of infants. Scien-tific American, 1966, 215, 80-92.

Brimble, A. R. The construction of a non-verbalintelligence test in Northern Rhodesia. Rhodes-Livingstone Journal, 1963, 34, 23-35.

Davenport, R. K., & Rogers, C. M. Perception ofphotographs by apes. Behavior, 1971, 39, 318-320.

Dennis, W. The human figure drawings of Bedouins.Journal of Social Psychology, 1960, 52, 209-219.

Deregowski, J. B. Difficulties in pictorial depth per-ception in Africa. British Journal of Psychology,1968, 59, 19S-204.

Farber, J. M. The effects of angular magnificationon the perception of rigid motion. (Doctoral dis-

sertation, Cornell University) Ann Arbor, Mich.:University Microfilms, 1972, No. 73-7134.

Ghent, L. Perception of overlapping and embeddedfigures by children of different ages. AmericanJournal of Psychology, 1956, 69, 575-587.

Gibson, E. J. Principles of perceptual learningand development. New York: Appleton-Century-Crofts, 1969.

Gibson, J. J. The visual world. Boston: HoughtonMifflin, 1950.

Gibson, J. J. What is a form? Psychological Review,• 1951, 58, 403-412.Gibson, J. J. The senses considered as perceptual

systems. Boston: Houghton Mifflin, 1966.Gibson, J. J. Three kinds of equivocal information

in line drawings. Ithaca, N.Y.: Author, September1969. (a) (Memorandum)

Gibson, J. J. The psychology of representation.Ithaca, N.Y.: Author, October 1969. (b) (Memo-randum)

Gibson, J. J. The information available in pictures.Leonardo, 1971, 4, 27-35.

Gibson, J. J., Olum, P., & Rosenblatt, F. Parallaxand perspective during aircraft landings. AmericanJournal of Psychology, 1955, 68, 372-385.

Gollin, E. S. Developmental studies of visual rec-ognition of incomplete objects. Perceptual andMotor Skills, 1960, 11, 289-298.

Gombrich, E. H. The story of art. London:Phaidon, 1971.

Goodman, N. Languages of art: An approach to atheory of symbols. Indianapolis: Bobbs-Merrill,1968.

Gregory, R. L. The intelligent eye. New York:McGraw-Hill, 1970.

Hay, J. C, Optical motions and space perception:An extension of Gibson's analysis. PsychologicalReview, 1966, 73, 550-565.

Hochberg, J. In the mind's eye. In R, N. Haber(Ed.), Contemporary theory and research in visualperception. New York: Holt, Rinehart & Winston,1968.

Hochberg, J. E., & Brooks, V. Pictorial recognitionas an unlearned ability: A study of one child'sperformance. American Journal of Psychology,1962, 75, 624-628.

Hudson, W. Pictorial depth perception in sub-cultural groups in Africa. Journal of Social Psy-chology, 1960, 52, 183-208.

Hudson, W. Cultural problems in pictorial percep-tion. South African Journal of Science, 1962, 58,189-195. (a)

Hudson, W. Pictorial perception and educationaladaptation in Africa. Psychologia Africana, 1962,9, 226-239. (b)

Hudson, W. The study of the problem of pietorialperception among unacculturated groups. Interna-tional Journal of Psychology, 1967, 2, 89-107.

Kennedy, J. M. Line representation and pictorialperception. Unpublished doctoral dissertation,Cornell University, 1971,

PICTURE. PERCEPTION 497

Kilbride, P. L., & Robbing, M. C. Linear perspective,pictorial depth perception and education amongthe Baganda. Perceptual and Motor Skills, 1968,27, 601-602.

Mooney, C. M. Age in the development of closureability in children. Canadian Journal of Psychol-ogy, 1957, 11, 219-227.

Mundy-Castle, A. C. Pictorial depth perception inGhanaian children. International Journal of Psy-chology, 1966, 1, 288-300.

Nadel, S. F. A field experiment in racial psychology.British Journal of Psychology, 1937, 28, 195-211;

Omari, I. M., & Cook, H. Differential cognitive cuesin pictorial depth perception. Journal of Cross-Cultural Psychology, 1972, 3, 321-325.

Pirenne, M. Optics, painting and photography.Cambridge, England: Cambridge University Press,1970.

Polanyi, M. Introduction. In M. Pirenne, Optics,painting and photography. Cambridge, England:Cambridge University Press, 1970.

Purdy, W. C. The hypothesis of psychophysical cor-respondence in space perception. (Doctoral dis-sertation, Cornell University) Ann Arbor, Mich.:University Microfilms, 1958, No. 58-5594).

Ryan, T. A., & Schwartz, C. Speed of perception asa function of mode of representation. AmericanJournal of Psychology, 1956, 69, 60-69.

Segall, M. H., Campbell, D. T., & Herskovits, M. J.The influence of culture on visual perception.Indianapolis: Bobbs-Merrill, 1966.

Shaw, R., Mclntyre, M., & Mace, W. M. The roleof symmetry in event perception. In R. B.MacLeod & H. Pick (Eds.), Studies in perception:Essays in honor of J. J. Gibson. New York:Cornell University Press, in press.

Smith, O. W. Comparison of apparent depth in aphotograph viewed from two distances. Perceptualand Motor Skills, 1958, 8, 79-81.

Smith, O. W., & Gruber, H. Perception of depth inphotographs. Perceptual and Motor Skills, 1958,8, 307-313.

Stern, W. Psychology of early childhood. NewYork: Holt, 1924.

Wilcox, B. L., & Teghtsoonian, M. The control ofrelative size by pictorial depth cues in children andadults. Journal of Experimental Child Psychology,1971, 11, 413-429.

Wohlwill, J. F. The perspective illusion: Perceivedsize and distance in fields varying suggested depth,in children and adults. Journal of ExperimentalPsychology, 1962, 64, 300-310.

Wolfflin, H. Principles of art history: The problemof the development of style in later art. (Trans.by M. D. Hottinger) New York: Dover, 1950.

Yonas, A., & Hagen, M. A. Effects of static andkinetic depth information on the perception of sizein children and adults. Journal of ExperimentalChild Psychology, 1973, 15, 254-265.

(Received August 10, 1973)

Documents

PICTURE PERCEPTION: TOWARD A THEORETICAL MODEL · 2016. 11. 7. · J. J. Gibson's new theory of picture perception is described, and a program ... usually by differences in color,