form and texture in hierarchically constructed...

Journal of Experimental Psychology:Human Perception and Performance1982, Vol. 8, No. 4, 521-535

Copyright 1982 by the American Psychological Association, Inc.0096-1523/82/0804-0521$(X>.75

Form and Texture in Hierarchically Constructed Patterns

Ruth Kimchi and Stephen E. PalmerUniversity of California, Berkeley

Perceived organization of hierarchically constructed patterns was investigatedthrough similarity judgments and a verbal description task. The number of ele-ments and their sizes relative to the configuration were varied in a series of fiveexperiments. The results show that in patterns composed of a few relatively largeelements, the elements are perceived as individual parts of the overall form andare perceptually salient. Increasing the number of elements and/or decreasingtheir size results in a perceived unified form associated with texture, representingthe structural properties of the elements as a group. In the latter case, theperceptual salience of the individual element decreases and the global form (orsometimes the texture) dominates perception. These findings suggest that theperceptual levels arising from global configuration of local elements may notcorrespond directly to these two geometrical levels in the stimulus domain asmuch previous work on "global versus local" processing has assumed. Rather,the mapping of the two independent geometrical levels into meaningful perceptuallevels depends critically on the number and relative size of the elements, thuschanging the perceived organization of the whole pattern.

An important perceptual issue concernsthe nature of the basic units of perception.In psychology this question can be tracedback to the controversy between two schoolsof perceptual thought: Structuralism andGestaltism. Structuralists have held that thebasic units of perception are independent lo-cal sensations that can form extended unitsonly through learning and experience. Ges-taltists, in contrast, have seen the primacyof extended units simply as a result of a cer-tain type of stimulus structure affecting anorganized perceptual system. A basic tenetof the Gestalt view of perception is that aspecific sensory whole is qualitatively differ-ent from the complex that one might predictby considering only its parts. The quality ofa part is determined by the whole in whichthis part is integrated. Thus, "local sensa-

This research was supported in part by National In-stitute of Mental Health Grant 1-RO1-MH33103-02to the second author and by National Institute of MentalHealth Grant 5-RO1-MH24316-03 to Eleanor Rosen.

We would like to thank Eleanor Rosch and DanielMorrow for stimulating discussions and helpful com-ments on this manuscript.

Requests for reprints should be sent to Ruth Kimchi,Department of Psychology, University of California,Berkeley, California 94720.

tions and all their relations will not have areal shape before it is segregated as a visualwhole" (KOhler, 1971, p. 162).

Although the basic flavor of the structur-alist approach has been retained in manycurrent models of perception (e.g., LaBerge,1973; Rumelhart & Siple, 1974; Treisman,1969), several theorists have recently pro-posed that perceptual processes proceedfrom global structuring toward more local,detailed, analysis (Broadbent, 1977; Navon,1977). The decomposition of a visual sceneis viewed as a hierarchical network of sub-scenes interrelated by spatial relationships(e.g., Palmer, 1975, 1977; Winston, 1975);it is claimed that higher order properties areprocessed first, followed by analysis of lowerorder properties.

This hypothesis has been experimentallytested by studying the perception of heir-archically constructed patterns in whichlarger figures are constructed by suitablearrangement of smaller figures. Throughoutthis article, we use the following terms todescribe the structure of such stimuli: Pat-tern refers to the entire stimulus figure, localelements to the smaller figures that make upthe pattern, and global configuration to thelarger figure constructed from the local ele-

521

522 RUTH KIMCHI AND STEPHEN E. PALMER

ments. Thus, global configuration does notrefer to the local elements at all, whereaspattern refers to both levels at once. We con-ceive of these two levels of patternstructure—global configuration and localelements—as geometrically independent.Whether or not they are perceptually inde-pendent is another matter.

By a set of converging operations, Navon(1977) demonstrated the perceptual priorityof global configurations. The main findingwas that conflicting information between thelocal and global levels (e.g., a large H madeup of small ss) had an inhibitory influenceon responding to the local letter in a Stroop-like interference task, but not to the globalletter. Navon interpreted these findings asevidence for the inevitability of global pre-cedence in visual perception.

However, studies done by Kinchla andWolf (1979), using a target search para-digm, Pomerantz and Sager (1975), usingGarner's (1974) sorting paradigm, McClean(1978) and Martin (1979), using the inter-ference paradigm, and Hoffman (1980), us-ing both the interference paradigm and thetarget search paradigm, seem to conflict withNavon's claims. Depending on variablessuch as the overall visual angle of the stimu-lus (Kinchla & Wolf, 1979), brightness(McClean, 1978), number and spacing ofthe local elements (Martin, 1979; Pomerantz& Sager, 1975), and clarity (distorted vs.undistorted forms; Hoffman, 1980), the localelement rather than the global configurationcan be favored perceptually.

Although differences in task variablesmay account for some inconsistencies in theresults of global-local studies, we questionsome assumptions underlying much of thisresearch with hierarchical stimuli. In par-ticular, it seems to have been frequently pre-supposed that there are two perceptual levelscorresponding directly to the global config-uration and local elements, and that the crit-ical question is which level gets processedfirst. The assumption that two independentperceptual levels correspond directly to twoindependent geometrical levels may be truein some cases and false in others, however.Thus, some clear notion of how hierarchicalpatterns are organized perceptually is an im-

portant prerequisite for asking meaningfulquestions about processing mechanisms.

We take as our starting point a phenom-enon discovered by Goldmeier (1936/1972)in studies of visual similarity. Using patternscomposed of different numbers and relativesizes of local elements, Goldmeier found thatproportional enlargement of a figure (thatis, an enlargement in which the sizes of boththe global configuration and the local ele-ments are increased equally by uniform di-lation) looked more similar to the originalfigure only when the original figure was com-posed of a few elements that were large rel-ative to the global configuration. When thefigure was composed of many elements and/or the elements were relatively small, a par-ticular sort of unproportional enlargementlooked more similar—one in which the globalconfiguration was enlarged but the mea-surements of the elements were unchanged.Goldmeier suggested that the latter type ofstimulus produces phenomenal separationbetween the overall form and the materialfrom which the form is constructed. Thisseparation disappears in the organization ofa pattern composed of a few relatively largeelements. When the figure is composed ofa few relatively large elements, they are per-ceived as individual parts of the overall form.When the figure is composed of many ele-ments and/or they are relatively small, theelements are relegated to the role of materialor, in other words, serve to define textureand do not interact with the form of thefigure. »

The distinction between patterns in whichthe elements are perceived as individualparts of the overall form rather than as tex-ture may resolve some of the inconsistenciesin the global-local studies. The distinctionalso may facilitate understanding the per-ception of the "local" and "global" aspectsof a pattern. Goldmeier's (1936/1972) worksuggests that the number and relative sizeof the elements are important factors in de-termining the role of the elements in a pat-tern. The present experiments investigate theeffects of these two factors on the perceptualorganization of patterns composed of ele-ments through similarity judgments and averbal description task.

FORM AND TEXTURE 523

Experiment 1

Experiment 1 parametrically investigatedthe phenomenon discovered by Goldmeier(1936/1972). Subjects were presented witha standard figure and two comparison fig-ures. All figures were composed of elements.The number of elements used to constructthe standard figure was varied. One com-parison figure was a proportional enlarge-ment of the standard figure (i.e., the overallfigure and the elements were proportionallyenlarged). The other comparison figure wasan unproportional enlargement having thesame overall size as the proportional enlarge-ment but containing elements whose size andspacing were the same as in the standardfigure. We expected that the probability ofchoosing the unproportional enlargementwould increase monotonically as the numberof elements increased due to greater sepa-ration between form and material.

Method

Subjects. Seven females and five males ages 21-36with normal vision volunteered to participate in the ex-periment.

Stimuli. There were 24 stimulus triads. Each stim-ulus triad was presented as three black patterns on awhite index card (22 cm X 13 cm). The triad containeda standard figure on the top and two comparison figuresside by side below the standard. Each pair of comparisonfigures consisted of a proportional enlargement and anunproportional enlargement of the standard figure. Oneset of the stimulus triads is shown in Figure 1. Theenlargement was in the ratio of 1:1.6. Half of the stim-ulus triads consisted of squares made up of smallsquares, and the other half consisted of triangles madeup of small triangles. In the standard figure there were1, 4, 9, 16, 25, and 36 elements for the squares and 1,3, 6, 10, IS, and 21 for the triangles. Subjects viewedthe cards from approximately 60 cm, although this var-ied from subject to subject in an uncontrolled fashion.At this distance, each individual element subtended .29°(degrees of visual angle); the global configurations sub-tended .29°, .95°, 1.34°, 1,53°, and 2.29° in the in-creasing-number conditions. In half of the stimulustriads for each type of figure the positions of the twocomparison figures were reversed.

Procedure. Subjects were asked to give their im-mediate and spontaneous impression about which of thetwo comparison figures looked more similar to the stan-dard figure. The subject sat in front of a table and sawone card at a time. Each subject gave a similarity judg-ment for each of the 24 cards. The order of the cardswas random over subjects.

Design. The three factors of the design were number

of elements (6 levels); figure (square, triangle); and sub-jects. All factors were combined orthogonally exceptthat the value at each level of the factor of number wasdifferent for the two types of figures.

Results and Discussion

Each response was scored as 1 if the un-proportional enlargement was chosen and as0 otherwise. Probabilities of choosing theunproportional enlargement as a function ofthe number of elements for each type of fig-ure are shown in Figure 2.

Since each subject gave two responses toeach combination of number and figure, ascore of 0, 1, or 2 was assigned to each sub-ject, and a three-way analysis of variance(ANOVA) was performed. The analysistreated the number factor as nested withinthe figure factor. The overall probability ofchoosing the unproportional enlargementwas significantly higher for the squares thanfor the triangles, F(l, 11) = 7.91, p < .01.The effect of number of elements was sig-nificant for both figures: for squares, F(5,55) = 25.41, p < .001; for triangles, F(5,55) = 7.10, p< .001.

A closer inspection of the data by binomialtests showed that when the number of ele-ments embedded in the square was 1 and 4,the probability of choosing the proportionalenlargement was significantly higher thanthe probability of choosing the unpropor-tional one (p < .001 and p < .05, respec-tively). When the number of elements was9 or more, the probability of choosing theunproportional enlargement was signifi-cantly higher (p < .01 for 9 and 16 elements;p < .001 for 25 and 36 elements). With thetriangles, the region of transition from pro-portional to unproportional enlargement waslarger: The probability of choosing the pro-portional enlargement was significantlyhigher for 1- and 3-element figures (p < .001and p < .01, respectively), whereas the op-posite holds only for 2!-element figures(p < .05).

The present experiment replicates Gold-meier's effect. When the figure was com-posed of a few relatively large elements, pro-portional enlargement was favored.Moreover, the response to the few-element


patterns was the same as to the 1-elementpattern, where there was no structural sep-aration between the local element and the

global configuration. When the number ofelements increased, the comparison figure inwhich the overall figure was enlarged but the

A A

A

A A

!••••••••••*•••••••*•••

Figure 1. The set of stimulus triads in Experiment 1. (Another 12 stimulus triads were identical to theones presented here, except that the position of the comparison figures was reversed.)


too

.75;

.501

• • SQUARES

*•--•* TRIANGLES

9 16 25

NUMBER OF ELEMENTS

Figure 2. Results of Experiment 1: The probabilities ofchposing the unproportional enlargement of the stan-dard figure as a function of the number of the elementsin the standard figure. (The error bars represent plusand minus 1 SE.)

measurements of the elements were keptunchanged (i.e., the unproportional enlarge-ment) was judged more similar to the stan-dard figure. Note that keeping the elements'measurements unchanged refers to both theelements and their interrelations (i.e., sizeand distance). Since both enlargements pre-serve the global configuration (transpositionin size does not affect the perceived shape),the switch from choosing the proportionalenlargement to choosing the unproportionalone can be interpreted as reflecting the ap-pearance of material (Goldmeier 1936/1972)or texture, a phenomenal quality represent-ing the structural properties of the elementsas a group rather than as individuated en-tities.

Thus, when the pattern is composed of afew relatively large elements, the elementsare perceived as individual parts of the over-all form. When the number of elements israther large, there are two phenomenal qual-ities: form and texture. Both are importantfor the impression of similarity. The elementloses its function as an individual part of theform and serves to define texture. Its prop-erties contribute to the emergent propertiesof the elements as a group—that is, texturalproperties such as density, regularity, andorientation—but are not represented specif-ically as figures in and of themselves.

The difference between squares and tri-angles regarding the appearance of theimpression of texture (as indicated by choos-ing the unproportional enlargement) mightbe due to the fact that the white spaces be-

tween the small triangles are themselves tri-angles (see Figure 1) and their number issmaller. If they are perceived as figure(rather than ground) the parametric differ-ence between the square and the triangleconditions diminishes.

It should be noted that since we usedforced-choice similarity judgments, the sim-ilarity judgments did not yield informationabout the degree of the similarity of a par-ticular comparison figure to the standard.However, all comparison figures were.pro-portional and unproportional enlargementsof the standard; the only manipulated factorwas the number of elements. Thus, the pres-ent approach was appropriate for studyingthe main question of interest in the presentstudy—namely, how increasing the numberof elements affects the choice between thetwo comparison figures. The same line ofreasoning also holds for the three experi-ments that follow.

Experiment 2

The interpretation of the previous findingsin terms of the function of the element in apattern has implications for the relative sa-lience of the local elements versus the globalconfiguration. When the pattern is composedof a few relatively large elements, the ele-ments are salient as figural parts of the over-all form. When the number of elements in-creases and they become part of the texture,the global configuration is more salient thanthe individual elements.

The second experiment tests this hypoth-esis. Subjects were presented with a standardfigure and two comparison figures, each con-sisting of a global square or triangle madeof local squares or triangles. In the same-configuration comparison pattern, differentelements were arranged in the same config-uration as the standard figure. In the same-element comparison pattern, the same ele-ments as in the standard figure were usedbut were arranged in a different configu-ration. The number of elements used to con-struct the standard figure was varied. If theelements are less salient when they are morenumerous and relatively smaller, then theprobability of perceiving the same-configu-ration pattern as more similar to the stan-


dard should increase as a function of thenumber of elements.

MethodSubjects. Twelve new subjects, eight females and

four males ages 2J-33 with normal vision, participatedin the experiment.

Stimuli. There were 16 stimulus triads. Each stim-ulus triad was presented as three black patterns on awhite card (15 cm X 13 cm) and contained a standardfigure and two comparison figures in the manner de-scribed under the heading Experiment 2. The set of thestimulus triads is shown in Figure 3. In half of the stim-ulus triads the standard figure was a "square" made upof either squares or triangles. In the other half the stan-dard figure was a "triangle" ma.de up of either trianglesor squares. In the standard figure there were 4, 9, 16,and 36 elements for the "squares," and 3, 10, 15, and36 for the "triangles." Subjects viewed the cards froma distance of approximately 60 cm. At this distance theglobal configuration subtended about 1.9° in the dif-ferent stimulus conditions. Each individual element sub-tended .76°, .48°, .38°, and .19°. Each type of com-parison figure appeared equally often in the right andleft position.

Procedure. The procedure was the same as in Ex-periment 1.

Design. The three factors of the design were numberof elements (4 levels); configuration (square, triangle);and subjects. All factors were combined orthogonally.(Technically, the number of elements were not exactlyequal for squares and triangles at Levels 2 and 3 of thatfactor, but they differed by only one element in eachcase.)


Each response was scored as a 1 if thesame-configuration stimuls was chosen and0 otherwise. The probabilities of choosingthe comparison figure having the same con-figuration are shown in Figure 4 as a func-tion of the number of elements embeddedin the standard figure.

Since each subject gave two responses toeach combination of number and configu-ration, the score 0, 1, or 2 was assigned toeach subject. A three-way ANOVA (Num-

AA

AAAAAAAAA • •••

AAAAAA AAA

AAAAAA

llii44**••••• AAAA

!•••••••»••

Figure 3. The complete set of stimulus triads in Experiment 2.


her X Configuration X Subjects) was per-formed. The difference between the"squares" and the "triangles" (F< 1) andthe Number X Configuration interaction(F< 1) were not significant. The effect ofthe number of elements was significant, F(3,33)= 10.82, p<. 001.

Closer inspection of the data by binomialtests showed that when the number of ele-ments was 3 (triangle configuration) or 4(square configuration), the probability ofchoosing the comparison figure having thesame elements arranged in a different con-figuration was significantly higher than theprobability of choosing the one having thesame configuration composed of differentelements (p < .01). When the number ofelements was 15 (triangle configuration) and36 (triangle and square configurations) theprobability of choosing the comparison fig-ure having the same configuration was sig-nificantly higher (p < .05).

The results of the present experiment sup-port our hypothesis. When the number ofelements increases, they are relegated to therole of defining texture (as indicated bychoosing the unproportional enlargement inExperiment 1); the global configuration isperceptually more salient than the individualelements (as indicated by the similarityjudgments in Experiment 2). Furthermore,the global configuration is phenomenally in-dependent of its local constituents: Replac-ing the elements of the figure by other ele-ments did not affect the perception of itsoverall form. As such the present experimentprovides a converging operation to define theseparation between overall form and thematerial from which the form is constructed.The finding that there was no local-to-globalinterference in a Stroop-like interferencetask in the case of inconsistency between theglobal shape and the shape of the local ele-ments (e.g., an H made of ss) for figurescomposed of many small elements (Kinchla& Wolf, 1979; Martin, 1979; Navon, 1977)is also consistent with this notion.

The results for the figures with a few rel-atively large elements showed the impor-tance of the individual elements for makingsimilarity judgments. However, the predom-inance of the elements does not necessarilyindicate that the similarity judgments were

.75

.50

.25

• • SQUARE CONFIGURATION

*•---* TRIANGLE CONFIGURATION

4 9 16 36

NUMBER OF ELEMENTS

Figure 4. Results of Experiment 2: The probabilities ofchoosing the comparison figure having the same config-uration as the standard figure, as a function of the num-ber of elements. (The error bars represent plus andminus 1 SE.)

based solely on the individual elements, sincethe actual global shape of the variant thatsupposedly had the same configuration wasdifferent. For example, the triangular con-figuration made of three relatively large tri-angles results in a "better" global trianglethan the triangular configuration made ofthree relatively large squares (see Figure 3).An unpublished study by Lotery (Note 1)attempted to overcome this difficulty by us-ing squares and rectangles with which it ispossible to obtain similar global configura-tions while varying the local elements. Thenext experiment replicated her study follow-ing the same experimental design as in thepresent experiment.

Experiment 3

Subjects were presented with a standardfigure and two comparison figures. All fig-ures consisted of a global square or rectanglemade of local squares or rectangles. The useof squares and rectangles made it possibleto minimize the confounding between theglobal outline and the local elements in pat-terns composed of a few relatively large ele-ments. Thus, we will get a clearer pictureof the importance of the individual elementsin such patterns for the impression of simi-larity. In addition, the elements and theirspatial arrangement were kept the samewhile varying the global configuration ofmany-element patterns. As a result globalconfiguration was pitted against texturerather than against the individual elementsas in the previous experiment. The question


KB:HiS

Figure 5. The set of figures from which the stimulus triads were constructed in Experiment 3.

was thus whether form or texture woulddominate the impression of similarity in themany-element conditions of the present ex-periment.

Method

Subjects. Twelve new subjects, six females and sixmales ages 24-30 with normal vision, participated inthe experiment.

Stimuli. There were 24 stimulus triads, constructedin the same manner as in Experiment 2. The set offigures from which the stimulus triads were constructedis shown in Figure 5. In half of the stimulus triads thestandard figure was a "square" made up of either'squares or rectangles. In the other half the standardfigure was a "rectangle" made up of either rectanglesor squares. In the standard figure there were 4, 16, and64 elements. Subjects viewed the cards from approxi-mately 60 cm. At this distance, the global square sub-tended 1.81° and the global rectangle subtended 3.81°in width and .80° in height. Each individual square ele-ment subtended .76°, .38°, and .19° in the increasingnumber conditions. Each individual rectangle elementsubtended 1.81° and .286°, .859° and .171°, and .38°and .057° in width and height respectively. In half ofthe stimulus triads for each type of figure the positionof the comparison figures was reversed.

Procedure. The procedure was the same as in theprevious experiments.

Design. The three factors of the design were numberof elements (3 levels); configuration (square, rectangle);and subjects. AH factors were combined orthogonally.


Each response was scored as a 1 if thesame-configuration stimulus was chosen and

0 otherwise. The probabilities of choosingthe comparison figure having the same con-figuration are shown in Figure 6 as a func-tion of the number of elements.

Since each subject gave four responses toeach combination of number and configu-ration, the score of 0, 1, 2, 3, or 4 was as-signed to each subject. A three-way ANOVA(Number X Configuration X; Subjects) wasperformed. The difference between the"squares" and the "rectangles" was not sig-nificant (F < 1), and the effect of numberdid not reach significance, F(2, 22) = 3.40,p > .05). The interaction between numberand configuration, however, was significant,F(2, 22)= 15.55, p<. 001.

Closer inspection of the data by binomialtests showed that when the standard figurewas either a square or a rectangle composedof 4 and 16 elements and a square composedof 64 elements, the probability of choosingthe comparison figure having the same ele-ments was significantly higher than the prob-ability of choosing the one having the sameconfiguration composed of different elements(p < .001 for 4 and 16 elements in squareand rectangle configurations;/) < .005 for 64elements in square configuration). When thestandard figure was a rectangle composedof 64 elements, performance reached chancelevel (p < .50)

The present results for few-element pat-terns converge with and clarify those of Ex-


periment 2. The preference for the compar-ison figure having the same elements as thestandard figure was also obtained when thechange in the global outline was minimizedwhile varying the local elements. Thus itseems fair to conclude that the individualelements in such patterns are perceptuallysalient as indicated by their importance forthe impression of similarity.

These results might suggest that in thecase of a pattern made of a few relativelylarge elements, the elements are percep-tually more salient than the global config-uration. However, to take the results of Ex-periment 1 into account as well (i.e., thepreference of the proportional enlargementin the case of pattern made of a few rela-tively large elements), an alternative inter-pretation seems more plausible—namely,that the elements are perceived as individualparts of the overall form. That is, there maybe no phenomenal separation between theoverall form and the elements.

At first glance the results for the many-elements patterns seem to contradict thehypothesis, supported by Experiment 2, thatthe form of many-element patterns is per-ceptually more salient than that of its indi-vidual elements. Increasing the number ofelements in a square configuration did notaffect the subjects' choices; although it hadsome effect in the expected direction whenthe standard figure was a rectangle, in bothcases the same-configuration comparisonfigure was not the preferred one. However,consideration of the particular constructionof the present stimuli shows that the presentresults do not necessarily contradict theabove hypothesis.

In Experiment 2, the spatial arrangementof elements changed when the global con-figuration was varied. For example, in asquare made up of squares, the square ele-ments were arranged in a gratinglike fash-ion. When the same elements were used toconstruct a triangle, the square elementswere arranged in a bricklike fashion (seeFigure 3). Thus, the global form was pittedagainst the shape of individual elements; theresults indicated the relative dominance ofthe global form. In Experiment 3, however,both the elements and their spatial arrange-ment were kept absolutely the same while

1.00

.75

.50

.25

I—• SQUARE CONFISCATION

>-—• RECTANGLE CONFIGURATION

_ _ _ -i

NUMBER OF ELEMENTS

Figure 6. Results of Experiment 3: The probabilies ofchoosing the comparison figure having the same config-uration as the standard figure, as a function of the num-ber of elements. (The error bars represent plus andminus 1 SE.)

varying the global configuration in many-element patterns (see Figure 5, rows 2-3).It has already been suggested that keepingthe elements and their measurements thesame results in an identical impression oftexture. Thus, in the many-element condi-tions of Experiment 3 the global form andthe texture were pitted against each other.The results can be interpreted as indicatingthat texture was more important for theimpression of similarity than the form, ex-cept when the standard figure was a rect-angle composed of 64 elements, where theform seems to compete with texture. Thestrong impression of the texture in Experi-ment 3 also was revealed in a postexperimentinquiry. Subjects described the standard fig-ure and the same-element comparison figurein the few-element conditions as having thesame element shape, whereas in the many-element conditions they were described asbeing like gratings or pieces cut from thesame cloth. The latter descriptions also sug-gest that textural properties rather than theindividual elements attracted the subjects'attention in many-element patterns.

The difference between many-elementrectangles and many-element squares thatsuggests the dominance of texture when thestandard figure was a square and a compe-tition between texture and form when thestandard figure was a rectangle might dem-onstrate another instance of asymmetry insimilarity judgments proposed by Tversky(1977) and Rosch (1975) (i.e., a rectangle


is more similar to a square than a square toa rectangle).

In Experiment 3 as well as in Experiment1, identical impressions of texture were ob-tained when the elements and their mea-surements were kept absolutely the same.Recent work on texture perception suggeststhat texture reflects crude information aboutthe structural properties of elements. Beck(1967; Beck & Ambler, 1972) demonstratedthat with respect to the dimension of shape,line-slope differences seem to be critical fortexture discrimination. Julesz (1975) pro-posed that texture perception is determinedonly by first- and second-order regularities,those that can be registered by the frequen-cies of points and of oriented dipoles thatconstitute the perception of density and"compactness." Other work done by Pom-erantz (1981) and Julesz (1981a, 1981b)suggests that although line-slope differencesand order-statistics differences are impor-

tant factors, they are not sufficient to explainall of texture discrimination. In any case, itis clear that current work on texture per-ception is consistent with our interpretationthat the same-element stimuli in Experiment2 differed in perceived texture, whereas thosein Experiment 3 did not.

Taken together, the results of Experi-ments 2 and 3 suggest that a pattern con-sisting of a few relatively large elements isencoded as individual units in various spatialrelationships to each other. When the num-ber of elements increases, the pattern is en-coded as a unified form associated with tex-ture, representing the structural propertiesof the elements as a group. While globalform of a many-element pattern is percep-tually more salient than the individual ele-ment (as the results of Experiment 2 indi-cate), it is not necessarily more salient thanthe impression of texture (as the results ofExperiment 3 indicate). Thus, "a change of

A AA A

A

A A

A A

A A

Figure 7. The set of figures from which the stimulus triads were constructed in Experiment 4.


form is by no means always more profoundthan a change of material" (Goldmeier,1936/1972, p, 52). It is, however, an inter-esting question to ask which factors deter-mine the relative dominance of form andtexture.

Experiment 4

For the stimuli used in Experiments 1, 2,and 3, the number of elements is correlatedwith their size relative to the global config-uration for strictly geometrical reasons: In-creasing the number of elements necessarilyresults in decreasing their relative size aslong as the spacing is kept constant. Withdiscrete elements, however, it is possible toseparate the effect of relative size from thatof number by constructing patterns in whichthere are only a few elements that are rel-atively small or large. The question of in-terest is whether the patterns with a fewsmall elements will produce the same resultsas those with a few large elements (indicat-ing the priority of number in the perceptualorganization of the pattern); those withmany small elements (indicating the priorityof size in the perceptual organization); orsomething else entirely.

MethodSubjects. Twelve new subjects, seven females and

five males ages 21-32 with normal vison, participatedin the experiment.

Stimuli. There were 16 stimulus triads, constructedin the same manner as in Experiment 2. The set offigures from which the stimulus triads were constructedis shown in Figure 7. The overall size of the figures(about 1.9°) and the number of elements were kept con-stant. There were four levels of relative size of the ele-ments, as shown in Figure 7. Each individual elementsubtended .76°, .48°, .38°, and .19° in the decreasinglevels of the size factor.

Procedure. The procedure was the same as in theprevious experiments.

Design. The three factors of the design were size ofelements (4 levels); figure (square, triangle); and sub-jects. All factors were combined orthogonally.


The scoring procedure was the same as inExperiment 2. The probabilities of choosingthe comparison figures having the same con-figuration as a function of the relative sizeof the elements embedded in the standardfigure are presented in Figure 8. A three-

1.00

.75

.50

.25

i SQUARE CONFIGURATION

. TRIANGLE CONFIGURATION

2.0 10 4.0 5.0 60 7.0

RELATIVE SIZE

6JD 9.0 10.0

Figure 8. Results of Experiment 4: The probabilities ofchoosing the comparison figure having the same config-uration as the standard figure, as a function of the- rel-ative size of the elements. (The relative size is repre-sented by configuration/element ratio. The error barsrepresent plus and minus 1 SE.)

way ANOVA (Size X Configuration X Sub-jects) was performed. The difference be-tween "squares" and "triangles" (F<1) andthe interaction between size and configura-tion (F < 1) were not significant. As ex-pected, however, the effect of relative sizeof the elements was significant, F(3, 33) =17.72, p<. 001.

Further analyses using binomial testsshowed that when the elements were rela-tively large (the first level), the probabilityof choosing the comparison figure having thesame elements arranged in a different con-figuration was significantly higher than theprobability of choosing the one having thesame configuration composed of differentelements (p < .01). When the relative sizeof the elements decreased, the probability ofchoosing the comparison figure having thesame configuration was significantly higher(p < .05 for the third level in square config-uration; p < .01 for the fourth level in bothsquare and triangle configuration).

The present results converge with thoseof Experiment 2. When the relative size ofthe local elements decreases, the individualelements become relatively less salient thanthe global configuration. The form becomesphenomenally independent of its elements:Replacing the elements by different elementsdid not affect the perception of the globalform. These results show that relative sizeof the element is a powerful factor in theseparation between form and the materialused to construct the form. Since relative


size covaried with number in the previousexperiments, it may have been the control-ling factor there, too. Unfortunately, we can-not isolate the effect of number from theeffect of relative size because the completeorthogonal design combining number andrelative size would require a geometricallyproblematic figure—a pattern composed ofmany relatively large elements.

Experiment 5

The relative salience of the local elementsand the global configuration of patterns wasfurther investigated by asking subjects tochoose the verbal description that best fit thepatterns among four alternatives. Two de-scriptions used the global configuration asthe grammatical subject and the local ele-ments as the grammatical object (e.g., "Atriangle .made of triangles") and the othertwo descriptions reversed their roles (e.g.,"Triangles arranged to form a triangle").The hypothesis was that subjects would pre-fer descriptions in which the grammaticalsalience of elements and configurations inthe sentence parallels their perceptual sa-lience. When there are many small elements,the global configuration is salient; subjectsshould choose a description in which the con-figuration is the grammatical subject. Whenthere are a few large elements, the local ele-ments are salient; subjects should choose adescription in which the elements are thegrammatical subject. Thus, we expected thepresent results to converge with those of theprevious experiments in indicating the dif-ferent roles of elements in a configurationfor different numbers and relative sizes ofelements.

MethodSubjects. One hundred twenty-one undergraduates

in a psychology course at the University of Californiaat Berkeley participated in the experiment.

Stimuli. Twelve patterns were used. The four pat-terns composed of a few relatively large elements in thepresent experiment were the four standard figures usedin Experiment 2 for the first level of number (see Figure3, row 1); the four many-element patterns in the presentexperiment were the four standard figures used in Ex-periment 2 for the fourth level of number (see Figure3, row 4); and the four patterns composed of a fewrelatively small elements were the four standard figuresused in Experiment 4 for the fourth level of relative size(see Figure 7, row 4).

Each pattern was presented on a separate page, with

four descriptions below each pattern. The descriptionsreferred to two aspects of the pattern: the global con-figuration and the element shape. In each descriptionone aspect was emphasized over the other by placingthe noun phrase that referred to it at the beginning ofthe description. Each of these two types of descriptionsappeared with and without an indication of the numberof elements in the pattern. For example, the four de-scriptions that appeared below a triangular configura-tion constructed from 36 triangles were (a) a trianglemade of triangles, (b) a triangle made of 36 triangles,(c) triangles arranged to form a triangle, and (d) 36triangles arranged to form a triangle. The order of pre-sentation of the patterns as well as the order of thedescriptions for each pattern were counterbalancedacross subjects.

Procedure. Subjects were instructed to mark thedescription that best fit his or her immediate impressionof the picture.

Results and DiscussionThe percentages of subjects who chose

each of the four alternative descriptions foreach type of pattern are presented inTable 1.

The results for many-element patternsand for a few relatively large element pat-terns are consistent with the results of Ex-periments 2 and 4. Sixty-six percent of thesubjects chose the descriptions that empha-sized the global form as the one that best fittheir impression of many-element patterns(p < .001). On the other hand, 70% of thesubjects chose the descriptions that empha-sized the local elements for few relativelylarge element patterns (p < .001).

Interestingly, the number of elements inmany-element patterns seemed irrelevant tothe description of the impression of the pat-tern, whereas with patterns of a few rela-tively large elements, the majority of sub-jects chose the description that emphasizedthe elements and referred to their number

Table 1Percentages of Subjects Choosing Each of theVerbal Descriptions for Each Type of Pattern

Pattern of elements

Verbal description

Global formGlobal form +

number of elementsElementsElements +

number of elements

36

60

628

6

3 or 4large

13

1711

59

3 or 4small

8

2210

60

FORM'AND TEXTURE 533

(e.g., "Three triangles arranged to form atriangle").

Another interesting finding was that theinconsistency between the global and locallevel of the pattern (e.g., a triangular con-figuration constructed from squares) had noeffect in the case of many-element patterns.This is just what would be expected if theglobal and local levels were perceptually sep-arate and independent. In the case of few-relatively-large-element patterns, however,more subjects chose the description that em-phasized the elements (80%) for "inconsis-tent" patterns than for "consistent" patterns(62%), although in both cases this descrip-tion was significantly favored over the de-scription that emphasized the global form.

The results for patterns composed of a fewrelatively small elements were somewhatsurprising. Although similarity judgmentsfor these patterns yielded the same resultsas for many-element patterns, the verbaldescription task yielded results similar tothose for patterns containing a few relativelylarge elements: 70% of the subjects chose thedescription emphasizing the elements ratherthan the global form. This finding might bedue to the nature of the task rather than toa perceptual effect. The perceptual salienceof the global configuration in such a patternwas clearly demonstrated through the sim-ilarity judgments in Experiment 4, where thepattern was presented in the context of twocomparison figures. The present task, how-ever, involved verbal description; it seemsplausible that conventions of conversation(communication) come into play. Whereasthe, global configuration might be easily per-ceived as a certain kind of form (e.g., a tri-angle or square), subjects may have beenaware that the form was "mentally con-structed," since no contours and no angleswere physically present. As a result, the de-scription that emphasized the global formmight be thought of as infelicitous (i.e., mis-leading a listener). Thus, although the globalform of a few-relatively-small-elements pat-tern is perceptually more salient than theindividual elements (as the similarity judg-ments indicated), the Gricean conversationalpostulate charging a speaker to tell the"truth" causes the subject to choose the de-'scription that emphasizes the elements(Grice, 1975). This interpretation is sup-

ported by a pilot study for the present ex-periment that required subjects to generatea description that best fit the configuration:Half of the subjects referred to these pat-terns as having an "imaginary" form.

General DiscussionThrough a set of converging operations,

the present experiments have demonstratedthat the perceived organization of hier-archically constructed patterns dependsstrongly on the number of local elements andtheir size relative to the global configuration.The primary theoretical questions are howto describe the phenomenological structurearising from such patterns and the percep-tual processes that produce this structure.

To begin, we repeat that we use the termsglobal configuration and local elements torefer to logically distinct levels of geomet-rical structure in the stimulus domain.Whenever small figures are positioned neareach other in such a way that their positionsform the pattern of a larger stimulus figure,these two geometrical levels are present, re-gardless of the number or relative size of theelements. Whether these two levels map intotwo independent perceptual levels is by nomeans clear. Unfortunately, many research-ers concerned with issues of "global" and"local" processing seem to assume that thereare always parallel perceptual levels corre-sponding to these geometrical levels. We willargue that this is only sometimes true.

On the psychological side, there do seemto be two rather distinct perceptual entities.Goldmeier (1936/1972) called them "form"and "material," and Julesz (1975) referredto them as "shape" and "texture." But theyare not to be equated with global and locallevels of stimulus structure as defined above.The present results suggest that the mappingof geometrical levels from the stimulus do-main into the perceptual levels of the psy-chological domain differs for patterns com-posed of many small elements and thosecomposed of a few large elements in the fol-lowing way: When many small elementscomprise a configuration, the local elementsare processed as texture and the global con-figuration as form. However, when few largeelements comprise a configuration, both thelocal and global levels are processed as form.The difference between these two cases re-


sides in the perceptual independence or de-pendence of the two geometrical levels.When local elements are processed as tex-ture, it is hypothesized that the global con-figuration is perceptually independent of thelocal elements and coded separately fromthem. When the local and global levels areboth processed as form, we suggest that thetwo are coded relative to each other.

This account attributes the "Goldmeiereffect" in Experiment 1 to the relative in-dependence/dependence of the configuraland elemental levels when the number andrelative size of elements varies (Goldmeier,1936/1972). Patterns composed of few largeelements are perceived to be more similarto proportional enlargements because theypreserve not only the global and local struc-ture but also the relationships between them.If both levels are coded relative to eachother, it seems reasonable that an enlarge-ment that does not distort the relationshipbetween the two levels would produce a moresimilar representation than one that doesdistort it. The situation is different for con-figurations composed of many small ele-ments, however, because the local level isrepresented independently of the global level.If the two are truly independent, in the senseof separate representations for form and tex-ture, then it would be expected that an en-largement whose textural representation wasidentical to the standard's would be per-ceived as more similar to it than an enlarge-ment whose textural representation differswith respect to size.

A perceptual theory that might accountfor the perceptual structure described abovehas recently been proposed by Julesz (1975,198la). Julesz has distinguished between afigure perception system and a texture per-ception system. Following this distinction itmight be the case that the two phenomenalqualities—form and texture—reflect the in-volvement of these two perceptual systems.The figure perception system operates on theglobal configuration, whereas the textureperception system operates on the entiregroup of local elements. Julesz (1975) hasalso suggested that when the elements in apattern are few and relatively large they re-sist becoming texture and are perceived bythe figure perception system. Another phe-nomenon that supports this suggestion is the

subitizing effect. A brief presentation (toavoid counting) of up to four clearly visibleobjects enables the observer to guess thenumber of objects without error, whereaswith more than seven objects, performancerapidly deteriorates.

An alternative account might be given interms of attentional mechanisms. It has beensuggested that visual attention can be drawnto a small area with high resolution or spreadover a wider area with some loss of detail(Eriksen & Hoffman, 1972; Treisman &Gelade, 1980). Suppose, then, that percep-tual attention is drawn primarily to the en-tire pattern. When the pattern is composedof many and/or relatively small elements,the individual element dissolves into theoverall characteristics of the design (i.e.,becomes part of the texture), and the globalconfiguration becomes the perceptual unitthat draws focal attention. To process theindividual elements, focal attention must beshifted to the local constituents. When theelements in the pattern are few and relativelylarge, they are less affected by the relativelylow resolution and the way is open for dif-ferent perceptual units to compete for per-ceptual dominance. The perceptual effectwill then depend strongly on intrinsic struc-tural properties of the pattern and its ele-ments, and on task demands.

A fully specific theory of perceptual mech-anisms underlying these phenomena mustawait more extensive empirical analysis.However, the results that elements are some-times perceived as individual parts of theoverall form and, at other times, as textureplaces constraints on perceptual processingmodels. For example, a model should notpredict the precedence of the global level ofstructure as a rigid perceptual law (e.g.,Navon, 1977). When the local elementsserve to define texture, perhaps being per-ceived by a different system (e.g., the textureperception system of Julesz, 1975), theglobal configuration can be perceived priorto, and independently of, the local form. In-deed, the perceptual priority of the globalform of a pattern composed of many rela-tively small elements, all else being equal,has been demonstrated in several studies(e.g. Navon, 1977; Martin, 1979; Robertson,1981). On the other hand, when the config-uration is constructed in an identical way


but the elements function as individual partsof the overall form (i.e., few relatively largeelements comprising the configuration), theprocessing of elements could even be priorto the perception of the overall form. A sim-ilar proposal has been suggested indepen-dently by Pomerantz (1981). Thus, any per-ceptual model will have to account for therelative perceptual separation of the config-ural and elemental levels when the numberand the relative size of the elements varies.

The present results on the perceptual or-ganization of hierarchical patterns suggestthat researchers should be clearer in theiruse of terms like global and local. Since therelationship between the geometrical andperceptual levels is not always the same, theterms should not be used to refer to bothdomains interchangeably. Rather, a clearterminology should be developed that re-flects the possibility of different sorts of map-pings from geometrical to perceptual levels.We suggest that global and local be reservedfor referring to levels of geometrical struc-ture in the stimulus (similarly for elementsand configurations), and that/orm (or shape)and texture (or material) be used to referto perceptual levels of subjective structure.Within this kind of framework, sensiblequestions about the correspondence betweenstimuli and percepts arising from them canbe posed clearly.

Reference Note1. Lotery, J. L. Global versus local precedence in visual

perception: Questioning the role of number of ele-ments. Unpublished manuscript, University of Cal-ifornia, Berkeley, 1980.

ReferencesBeck, J. Perceptual grouping produced by line figures.

Perception & Psychophysics, 1967, 2, 491-495.Beck, J., & Ambler, B. Discriminability of differences

in line slope and in line arrangement as a function ofmask delay. Perception & Psychophysics, 1972, 12,33-38.

Broadbent, D. E. The hidden preattentive processes.American Psychologist. 1977,32. 109-118.

Eriksen, C. W. W., & Hoffman, J. E. Temporal andspatial characteristics of selective encoding from vi-sual displays. Perception & Psychophysics, 1972,12,201-204.

Garner, W. R. The processing of information and struc-ture. Hillsdale, N.J.: Erlbaum, 1974.

Goldmeier, E. Similarity in visually perceived forms.Psychological Issues. 1972, 8(1, Whole No. 29).(Originally published, 1936)

Grice, H. P. Logic and conversation. In P. Cole &

J. I. Morgan (Eds.), Syntax and semantics. Vol. 3:Speech acts. New York: Seminar Press, 1975.

Hoffman, J. E. Interaction between global and locallevels of a form. Journal of Experimental Psychol-ogy: Human Perception and Performance, 1980, 6,222-234.

Julesz, B. Experiments in the visual perception of tex-ture. Scientific American, 1975, 232, 34-43.

Julsez, B. Figure and ground perception in briefly pre-sented isodipole textures. In M. Kubovy & J. R. Pom-erantz (Eds.), Perceptual organization. Hillsdale,N.J.: Erlbaum, 1981. (a)

Julesz, B. Textons, the elements of texture perception,and their interaction. Nature, 1981, 290, 91-97. (b)

Kinchla, R. A., & Wolf, J. M. The order of visual pro-cessing: "Top-down," "bottom-up" or "middle-out."Perception & Psychophysics, 1979, 25, 225-231.

KOhler, W. Human perception. In M. Henle (Ed.), Theselected papers of Wolfgang Ko'hler. New York: Liv-eright, 1971.

LaBerge, D. Attention and the measurment of percep-tual learning. Memory & Cognition, 1973, /, 268-276.

Martin, M. Local and global processing: The role ofsparsity. Memory & Cognition. 1979, 7, 476-484.

McClean, J. D. Perspectives on the forest and trees:The precedence of parts and whole in visual pro-cessing. Unpublished doctoral dissertation, Universityof Oregon, 1978.

Navon, D. Forest before trees: The precedence of globalfeatures in visual perception. Cognitive Psychology,1977, 9, 441-474.

Palmer, S. E. Visual perception and world knowledge.In D. A. Norman & D. E. Rumelhart (Eds.), Ex-plorations in cognition. San Francisco: Freeman,1975.

Palmer, S. E. Hierarchical structure in perceptual rep-resentation. Cognitive Psychology, 1977, 9, 441-474.

Pomerantz, J. R. Perceptual organization in informationprocessing. In M. Kubovy & J. R. Pomerantz (Eds.),Peceptual organization. Hillsdale, N.J.: Erlbaum,1981.

Pomerantz, J. R., & Sager, I. C. Asymmetric integral-ity with dimentions of visual perception. Perception& Psychophysics. 1975, 18, 460-466.

Robertson, L. R. Global processes and the perceptionof disoriented figures (Doctoral dissertation, Univer-sity of California, Berkeley, 1981). Dissertation Ab-stracts International, 1981, 42. 147A.

Rosch, E. Cognitive reference points. Cognitive Psy-chology. 1975, 7, 532-547.

Rumelhart, D. E., & Siple, P. A. Process of recognizingtachistoscopically presented words. PsychologicalReview, 1914,81, 99-118.

Treisman, A. Strategies and models of selective atten-tion. Psychological Review, 1969, 76, 282-299.

Treisman, A., & Gelade, G. A feature-integration the-ory of attention. Cognitive Psychology, 1980,12, 97-136.

Tversky, A. Features of similarity. Psychological Re-view, 1977, 84, 327-352.

Winston, P. H. Learning structural descriptions fromexamples. In P. H. Winston (Ed.), The psychologyof computer vision. New York: McGraw-Hill, 1975.

Received November 23, 1981Revision received December 24, 1981 •

form and texture in hierarchically constructed...

Documents