UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Completion of occluded objects in early vision : an exploration of spatial limits Shore, David I. 1993

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_1993_fall_shore_david.pdf [ 2.62MB ]
Metadata
JSON: 831-1.0086297.json
JSON-LD: 831-1.0086297-ld.json
RDF/XML (Pretty): 831-1.0086297-rdf.xml
RDF/JSON: 831-1.0086297-rdf.json
Turtle: 831-1.0086297-turtle.txt
N-Triples: 831-1.0086297-rdf-ntriples.txt
Original Record: 831-1.0086297-source.json
Full Text
831-1.0086297-fulltext.txt
Citation
831-1.0086297.ris

Full Text

We accept this thesis as conformingto the required standardDr. J.T. EnnsDr. L.M. WardDr. C. BlahaCompletion of Occluded Objects In Early Vision:An Exploration of Spatial LimitsbyDavid I. ShoreB.Sc., McMaster University, 1991A THESIS SUBMITTED IN PARTIAL FULFILMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF ARTSinTHE FACULTY OF GRADUATE STUDIESDEPARTMENT OF PSYCHOLOGYTHE UNIVERSITY OF BRITISH COLUMBIAAugust, 1993© DAVID I. SHORE, 1993In presenting this thesis in partial fulfilment of the requirements for an advanceddegree at the University of British Columbia, I agree that the Library shall make itfreely available for reference and study. I further agree that permission for extensivecopying of this thesis for scholarly purposes may be granted by the head of mydepartment or by his or her representatives. It is understood that copying orpublication of this thesis for financial gain shall not be allowed without my writtenpermission.(Signature) Department of 25 (1 C' L€( 5YThe University of British ColumbiaVancouver, CanadaDate  ocf- 1 ‘1-, (4 q3 DE-6 (2/88)AbstractOur visual experience is of complete objects despite the fact that theretina is often given only partial views of these objects. Objectivesupport for this perceptual completion of partly occluded objects wasgenerated in Experiment 1 in which same-different response time (RT)was measured for objects that were either complete, partly occluded,or notched.^Complete and partly occluded objects yielded similar RTs,while notched objects took much longer to match. For larger amountsof overlap the occluded condition took longer than the whole condition.This suggested that rapid completion is limited to relatively small gapsizes. In Experiment 2, we varied the amount of occlusionsystematically in order to find the limit of rapid completion. With 25%object occlusion or less, we again found evidence for rapid completion;beyond this point subjects appeared to use a strategy that was similarto that used in the notch condition. The separability of completionfrom other cognitive processes was investigated using two strategies.First, in Experiment 3 the task demands were varied from anidentification task to a categorization task.^Similar spatiallimitations were found for both of these tasks. Second, in Experiment4 the target was a moving object and it was again found thatcompletion was separable from other later processes such that movingtargets were easier to complete but no easier to classify. Theseresults support the proposal of a preattentive completion process thatis spatially limited.Table of contents^Abstract   iiTable of contents  ^iiiList of Tables  ^vList of figures  ^viAcknowledgments ^  viiIntroduction^ 1Experiment 1  13Method ^  16Subjects ^  16timuli   16Apparatus  17Procedure ^  18Results ^  20Discussion  24Experiment 2a ^  25Method ^  27Subjects  27timuli   27Apparatus ^  28Procedure  28Results ^  29Discussion   31Experiment 2b ^  32Method    32Subjects  32timuli   33Apparatus ^  33Procedure.   33Results ^  33Discussion ^  36Experiment 3  38Method ^  40Subjects ^  40^timuli   40Apparatus  40Procedure ^  40Results ^  41Discussion  44Experiment 4 ^  45Method  46Subjects  46timuli   48Apparatus ^  49Procedure  49Results ^  49Discussion  54General Discussion ^  55References ^  63Appendix A - Raw Tabled Data ^  66ivList of TablesTable 1. Results of Experiment 1: Raw Data (RT and Error) ^ 6 7Table 2. Results of Experiment 2a: Raw Data (RT and Error) ^ 68Table 3. Results of Experiment 2b: Raw Data (RT and Error) ^ 6 9Table 4. Results of Experiment 3: Raw Data (RT and Error) ^ 70Table 5. Results of Experiment 4 - Static Conditions:Raw Data (RT and Error) ^  71Table 6. Results of Experiment 4 - All Move Conditions:Raw Data (RT and Error) ^  7 2Table 7. Results of Experiment 4 - Target Only Move Conditions:Raw Data (RT and Error) ^  73List of figuresFigure 1 Phenomenal Demonstration of Early and Late Completion^3Figure 2. Sample Stimuli and Naming System  ^5Figure 3. Example of Line Fragmentation Task (Enns & Rensink, in press) 7Figure 4. Example of Notched Task (Enns & Rensink, in press) ^ 10Figure 5. Sample Stimuli in Experiment 1 ^  14Figure 6. Results in Experiment 1 ^  22Figure 7. Results In Experiment 2a  30Figure 8. Results in Experiment 2b ^  35Figure 9. Results in Experiment 3  42Figure 10. Sample stimuli used in Experiment 4 ^  47Figure 11. Results in Experiment 4 ^  53viAcknowledgmentsFirst and foremost, I would like to thank Dr. Enns for his support,both mental and material, in this project. From the very beginning hemade me feel comfortable and welcome. As well, I would like to thank mycommittee members for their suggestions and patience; especiallyLawrence Ward for his seemingly unending knowledge base andwillingness to share. To the Enns' lab members who have made my workenvironment a happy and conducive one I extend my warmest thanks: DebAks for her mentorship in the ways of being a Grad student (what a goodinfluence you were); Darlene Brodeur for her Tye Dye Parties and thepatient way she answers all my questions; Lana Trick for her suggestions,and all the coffee we drank; Tracy Wood, for all her help in getting thingsdone, running some of my subjects and for just being around to brightenup the lab. I also wanted to recognize the contribution of Geoff Small forhis participation in running the first experiment.On a more personal note I owe perhaps my greatest thanks to TrudyMohrhardt for her endless devotion and constant support. Other friendsthat need mention include Matt Pollack, Jon and Devin Summerhayes fortheir editing suggestions; Zonya, Matt, Wilf and Gaia for allowing Trudyand I to live with them for so long; and David Tessler who has introducedme to more B.C. wonders than I can remember. Finally I want to thank myfamily members who have supported me throughout my education; Barb,Abe, Pearl, and Elan. And a special thanks to Dave Shore (the other one)for introducing me to all of my Vancouver Friends.viiEarly Completion1IntroductionWe experience the visual world as a collection of whole objectsdespite the fact that the retina receives only fragmentary informationabout them. For instance, when distant objects are partly occluded bynearer objects we nevertheless perceive the more distant objects aswhole. This phenomenological observation leads to questions such as:"How does the visual system go beyond the information given to generatethe experience of a whole object?", "How early in the visual stream isthis interpolation conducted?" and, "Are there limitations in thetemporal and spatial extent of this proposed process?". In this study Iwill show that there are spatial limits to the completion process. Theselimitations follow logically from the claim that object completion is anearly vision mechanism (Enns, 1992; Zucker, 1987).Kanizsa (1979) describes two perceptual processes we use to gobeyond the fragmentary information on the retina. The primary one"consists of an organization of proximal stimulation (Kanizsa, 1979,pp5)" which includes the completion of partly occluded objects usingsimple gestalt rules. The secondary process acts on the informationsupplied by the first to make perceptual inferences and behavioraldecisions. The validity of this distinction can be observed whenEarly Completion2comparing Figure 1 a and 1 b, which are both representations of the Neckercube (Figure 1c). In Figure 1 a it is very difficult to perceive the entirecube and even more difficult to experience the depth reversal for whichthe Necker cube is famous. If one is able to identify Figure 1 a as a Neckercube it is presumably because the presented fragments can be integratedinto a somewhat complete whole based on past experience. In Figure 1 bthere is no requirement for such inferences because the object iscompleted by a primary process and presented to consciousness as acomplete Necker Cube (Figure lc). Thus, the visual system can bedichotomized into two processes: a primary one which can organize theproximal stimulus into larger wholes following Gestalt like rules, and asecondary one that can reconstruct and recognize objects based onprevious knowledge and the information passed on from the primaryprocess. This information may, at times, be nothing more than thefragmentary information presented on the retina.In proposing this primary/secondary dichotomy, researchersgenerally assume that primary completion processes are active early inthe visual stream by a fast, automatic and spatially parallel mechanism(Enns, 1992; Enns & Rensink, 1993; Enns & Rensink, in press; Kanizsa &Gerbino, 1982; Kellman & Shipley, 1992; Ramachandran, 1992, 1993).Early Completion3ABCFigure 1. Phenomenal Demonstration of Early andLate CompletionEarly Completion4Much of the previous research addressing the question of objectcompletion has used phenomenological methodologies or appeals tosubjective impressions (Kanizsa, 1979; Kellman & Shipley, 1992). Morerecent efforts, including the current work, have used performance basedmeasures to study the completion mechanism in an objective fashion(Enns, 1992; Enns & Rensink, in press; Gerbino & Salmaso, 1987; Sekular &Gaber, 1993; Sekular & Palmer, 1992). Before describing theseprocedures and findings, I will discuss the naming of stimuli, since noconsistent naming practices can be found in the literature.For the most part, discussion will be limited to the first threeconditions shown in Figure 2. In the whole condition (Figure 2a) a targetitem is presented in its entirety either on top of another object or byitself. In the occluded condition (Figure 2b), the target object ispresented behind an irrelevant occluder. In the notched condition (Figure2c), the target object has the same outline as the occluded object withthe exception that there is no visible occluder. Thus a retinal differencebetween the whole and occluded objects is the occluded region of theoccluded object. Similarly, an important difference between the notchedand occluded objects is the presence of the occluder in the notch for theoccluded condition.Condition NameA) whole(TPC)completeB) occluded(PC)Experiment 1(Gerbino and Salmaso, 1987)Experiments2, 3, and 4C) notched(C)incompleteD) occluded-truncated(PC.t)E) whole-truncated N/AEarly Completion5Figure 2. Sample stimuli and naming systemEarly Completion6The remaining conditions are used to control for several possibleconfounds. For example, reaction time differences between the notchedand occluded objects may be due to the presence of the concavity on thenotched object (See Figure 2). To control for this interpretation, occludedand whole conditions are often run with notched distractors so that thesame junction information is present. The occluded-truncated (Figure 2d)and the whole-truncated (Figure 2e) conditions are examples of this typeof control figure. For the remainder of this paper I will make exclusiveuse of these terms.Several experiments have been conducted to investigate the ability ofearly vision to complete objects behind perceived occluders. Enns andRensink (in press; Enns, 1992) have looked at two types of occlusionsituations: one involving the completion of simple line segments across afilled or empty gap, and another looking at the completion of 2-Dgeometrical shapes that have a portion missing. In the first case the twofragments can be considered as separate entities or as part of a commonobject. In the second case there is only one fragment that simply has aportion missing. Both of these lines of research have used the visualsearch task (Treisman & Galade, 1980).Early Completion7Figure 3. Example of line fragmentation task(Enns & Rensink, in press).Early Completion8In the line fragmentation task, the attribute that distinguished thetarget from the distractor was the location of the gap, either in thecentre of the line segment (distractors) or offset to either side (target,See Figure 3). The subjects task was to indicate if a target item waspresent or not. When the gaps were unfilled (Figure 3a) the task was veryeasy (search slopes averaged 6 ms per item for target present and 4 msper item for target absent) as subjects could use the length of the visiblefragments as features on which to base their search. When the gaps werefilled with a 2-d occluder (figure 3b) the task became more difficult (29ms and 55 ms per item for target present and absent respectively),presumably because subjects were completing behind the objects and thusthe target and distractors did not differ in overall length. When a 3-doccluder was present in the gaps (Figure 3c) the task became even moredifficult (47 ms and 88 ms per item) indicating that completion was evenmore efficient. The fact that search was difficult when an occluder waspresent but easy when no occluder was present supports the claim thatcompletion behind an occluder occurs preattentively. If subjects did notcomplete both the target and distractor line segments very rapidly and inparallel then the results for the second two experiments would look morelike the first experiment. Finding steeper search slopes indicates thatEarly Completion9completion is occurring before a parallel search can be conducted. Notethat this experiment and the following one directly support the proposalthat completion occurs automatically (i.e. without voluntary control)because subjects complete even though it is to their advantage not to.The second set of figures used by Enns and Rensink (in press) is moresimilar to the stimuli used in the current investigation (Figure 4). In thistask, the subject searched for a notched black square. The distractorswere whole black squares. When the notch was empty (Figure 4a) andcould thus serve as a feature for which to search, the task was very easy(7 ms and 8 ms per item). On the other hand, when the gap was filled by avisible occluder (Figure 4b), search was very difficult (36 ms and 66 msper item). The retinal stimulation from the targets in these twosituations was identical. The difference in reaction time slopes betweenthe conditions can be explained by the automatic completion of theoccluded gap such that the representation of the target object was thesame as the distractor items (whole squares) and thus required attentionto locate. This experiment again demonstrates that completion occursbefore a parallel search can be conducted (i.e. early in vision).ABEarly Completion10Figure 4. Example of notched task(Enns & Rensink, in press).Early Completion11The rapid nature of the completion process was investigated bySekular and Palmer (1992) using a high speed priming paradigm. Theyprimed subjects with a whole, occluded, or notched circle or square, thenrequired subjects to make a same-different response to a pair ofcomplete or incomplete items in the probe. The stimulus onset asyncrony(SOA) between the prime and probe was varied systematically between 10ms and 400ms. When the prime was a whole or notched figure, subjectswere faster at matching the corresponding complete or incompletefigures regardless of SOA. When the prime was an occluded figure therewas an interesting interaction with SOA. Before an SOA of 200 ms,subjects were faster to match a pair of incomplete items whereas after200 ms they were faster on the complete items. This indicates thatbefore 200 ms, subjects still had a fragmented representation of theoccluded figure, while after 200ms this representation was completed.One final experiment in the literature speaks to the rapid nature ofcompletion and follows more directly from the phenomenal demonstrationpresented earlier in the introduction. Gerbino and Salmaso (1987) hadsubjects match whole shapes with comparison shapes that were eitherwhole, occluded, or notched. They found that reaction times to match theoccluded and whole shapes did not differ whereas reaction times for theEarly Completion12notched condition were significantly longer. This was taken to indicatethat the notched shapes were being matched by a different, slower andlater process, perhaps one corresponding to Kanizsa's secondary process.The occluded objects were completed using the primary process and thenmatched to the whole object. Presumably all of these objects are matchedusing some attention-based operation and the advantage found for thewhole and occluded objects reflects an easier feature matching processthan that needed for the notched figures.In summary, it seems that partly occluded objects can be completedrapidly, automatically, and spatially in parallel.^These characteristicsdefine the completion of occluded objects as a preattentive process. Oneimplication of the spatially parallel nature of completion is the existenceof local operators of limited spatial extent (Enns, 1992; Enns & Rensink,1993; Zucker, 1987). For completion to be conducted across the scene inparallel each local area must be processed independently, implying thatthe proposed operator must have a limited neighbourhood of operation.When the area of occlusion does not fall within this area, earlycompletion should not be successful and the fragmented version presentedto the retina would be passed on to later visual routines foridentification. The purpose of this study is to confirm that the proposedEarly Completion13local operator is indeed spatially limited.^It was predicted that forlarger gap sizes than those sampled in the literature, there would be afailure of the completion process within early vision. To test thisprediction, a replication of the Gerbino and Salmaso (1987) experimentwas conducted for two gap sizes.Experiment 1The first experiment had two purposes: to replicate the Gerbino andSalmaso (1987) result using our current apparatus, and to determine ifthe results associated with occluded objects having small degrees ofoverlap would generalize to objects with larger gap sizes. There weretwo reasons for examining the larger gap sizes. First, all of the previousexperiments in the literature used objects that were only occluded to avery small extent (i.e. at least 75% of the object was visible) and thislimitation has theoretical implications for the completion process.Second, just as Sekular and Palmer (1992) showed that the proposedcompletion process is temporally limited, we need to discern if it is alsospatially limited, as is predicted by the characteristics of preattentiveprocesses. To these ends, the Gerbino and Salmaso (1987) study wasreplicated for two levels of occlusion. The small gap size was comparableto the original stimuli, whereas the large gap stimuli had about one thirdEarly Completion14Condition Name Small Gap Size Large Gap SizeA)whole(TPC)complete4WAAlk AbalB)occluded(PC)Alk, AA,C) notched(C)incompleteotAikFigure 5. Sample stimuli for Experiment 1Early Completion15more of the object covered. Examples of the stimuli used can be seen inFigure 5.It was predicted that for the small gap size, like in the Gerbino andSalmaso (1987) study, the whole and occluded stimuli should not differ interms of reaction time or accuracy of matching. On the other hand, thenotched condition should show longer reaction times and decreasedaccuracy in comparison to both the whole and the occluded conditions.For the large gap size, several patterns of results were possibledepending on the nature of the completion process. If completion isgeneral and can complete any gap size quickly then the same pattern ofresults might be expected. Alternatively, if completion is spatiallylimited, the occluded condition would be more difficult to match than thewhole condition for the large gap size only if indeed the large gapexceeded the size of the local operator field.There are two operational definitions that will be used to show thatsubjects are using the early vision strategy associated with thecompletion process. The first concerns the difference, in terms of RT andaccuracy, between the whole condition and the occluded condition. Ifcompletion occurs early in processing, then these two stimuli should beperceived as the same and not show any differences in reaction time. TheEarly Completion16second operational definition pertains to the notched condition. Since theretinal stimulation, in terms of the target item, is identical in occludedand notched conditions, the subjects' RT and accuracy should also showsimilar patterns. To the extent that the occluded condition is faster thanthe notched condition we can infer that early object completion hasoccurred.Method Subjects. Ten volunteer subjects were drawn from the undergraduatesubject pool at the University of British Columbia. In return forparticipating, subjects received extra credit in a psychologyundergraduate course. There were five right handed males, four righthanded females and one left handed female. All subjects had normal orcorrected to normal vision.Stimuli. A display consisted of a single target item displayed 7°above or below fixation and a comparison configuration consisting ofeither one or two items displayed opposite the target item (also 7° aboveor below fixation). Individual items consisted of diamonds (2.9° x 2.9°),triangles (3.8°x1.9°), and hexagons (3.5°x1.8°). The comparison itemscould be spaced at one of three distances. In the small gap condition, theoverlap was about 0.3° of visual angle (comparable to the stimuli used inEarly Completion17Gerbino & Salmaso, 1987). For the large gap, this value was about 1.2° ofvisual angle. In the control condition, there was no overlap and there waseither one or two items (0.25° apart). The target could have one of thefive relations to the comparison item. The first four possiblecombinations were given in the introduction under naming systems (SeeFigure 2 and page 4). In addition to these four conditions (whole,occluded, occluded-t, notched) 50% of the trials required a differentresponse. In these trials the comparison items were different from thetarget item.In total there were fourteen stimulus configurations. Tencorresponded to the cells in a two by five factorial design where the twofactors were gap size (small and large) and condition (whole, occluded-c,occluded-t, notched, and different). The remaining four trial types werecontrol conditions. There were two types of control conditions: single anddouble. In the single condition there was only one comparison item whichwas either the same as or different from the target item. In the doublecondition there were two comparison items. These items could eitherboth be different than the target or one of them could be the same as thetarget. See Figure 5 for examples of the whole, occluded, and notchedconditions.Early Completion18Apparatus. Subjects responded using a standard Macintosh (Mac)keyboard which was connected to a Mac plus. The stimuli were presentedon the Mac plus screen which had the contrast and brightness set to thehighest level. VScope software (Rensink & Enns, 1992) was used togenerate the stimuli and record millisecond reaction times and errorrates.Procedure. When subjects arrived they were seated in front of acomputer and asked to fill out two forms. The first was a simple sign-insheet. The second consisted of an instruction sheet with consent formattached and a biographical questionnaire asking about age, handedness,sex, eyesight, and year of university. Subjects were told to read theinstructions carefully and sign the consent form if they agreed toparticipate. The instructions emphasized equally the importance ofaccuracy and speed of responding. After the subjects signed the consentform they were asked if they had any questions. Following theseintroductory procedures the experimenter emphasized the importantpoints of the task. First, subjects were to try to keep their eyes fixatedon the centre of the screen throughout the experiment. Second, the factthat some of the items would be truncated was discussed and it wasEarly Completion19emphasized that these items should be treated as exemplars of thecomplete form. Third, the response keys to be used were reviewed.Subjects were then given at least 15 trials of practice. Theypracticed until they could perform 5 consecutive trials correctly. If thesubject was having difficulty, the instructions were reviewed again andthey were prompted for any questions.Subjects sat with their faces approximately 40 cm from the screenand were free to move. The task was to indicate if one of the comparisonitems was the same as the target item. They used the index finger of onehand on the keyboard to indicate same and the index finger of the otherhand to indicate different. The side of responding was counterbalancedacross subjects.Each block began by prompting the subject to press any key to begin.When subjects signalled they were ready, the first display appeared after750ms. The display lasted for 200ms. After the subject responded orafter three seconds had elapsed a feedback symbol was presented for915ms. This served as the fixation for the next trial which was presentedafter a 750ms inter-trial-interval during which the screen was blank.Feedback consisted of either a plus sign, a minus sign, or an opencircle indicating that the preceding response was either correct,Early Completion20incorrect, or beyond the three second limit. As well, at the end of eachblock the total percent error for that block was presented to the subject.There were 10 blocks of 60 trials each. The subject was encouragedto take self-paced breaks between blocks. The experimenter remained inthe room during the practice trials to answer any questions and monitorfor the criterion number of correct responses. During the actualexperiment the experimenter left the room but was available if thesubject had questions or problems.Results The first block of trials and the first three trials of every subsequentblock were treated as practice and not included in the analysis. This left504 trials to be analyzed for each subject. For these trials the error ratewas calculated for each condition by dividing the number of incorrecttrials by the total number of trials for that condition. The mean reactiontimes for the remaining correct trials was also calculated for eachsubject. These mean reaction times and error rates were submitted to arepeated measures ANOVA. Follow up analyses and a priori predictionswere conducted using either a simple main effects ANOVA forEarly Completion21interactions or the Fisher LSD procedure for pairwise comparisons. TheMS e used was always taken from the highest order interaction of theoverall analysis. Sphericity assumptions were tested using both theGreenhouse & Geiser (1959) technique and the Huynh & Felt (1970)technique. Throughout the four experiments there were no seriousviolations of the sphericity assumption. Whenever there was a violation,the lower probability value will be presented.The same trials were submitted to a two by four repeated measuresANOVA where the two factors were aap size (small and large) andcondition (whole, occ, occ-t, notch). The main effect of gap size wassignificant [F(1,9)=5.3, MS e =6909.7, p<.05] as was the effect of condition[E(3,27)=35.9, MSe=6212.5, p<.0001]. These differences were moderated bya significant interaction [F(3,27)=10.9, MS e=2486.7, p<.0001]. As can beseen in Figure 6, this interaction was due primarily to the significantdifference between the occluded trials and the whole trials for the largegap size [t(27)=-5.9, p<.0001], but not for the small gap size [t(27)=-0.7].The occluded trials were faster in the small gap condition than in thelarge gap condition [t(27)=-4.7, p<.0001]. The same was true for thenotched trials [t(27)=-4.8, p<.0001]. As well, the occluded trials weregenerally faster than the notch trials [t(27)=-6.1, p<.0001].Early completion22Figure 6. Results of Experiment 1Early Completion23In terms of errors, the main effect of condition was significant[F(3,27)=25.5, MS e=.012, p<.0001] while that for gap size was not[E(1,9)=4.3, MS e=0.011]. These differences were again moderated by asignificant interaction [F(3,27)=4.1, MS e=.009, p<.05]. Although the errorsfollow the same pattern as the reaction times, the only significantdifferences were found in the notch conditions which were more errorprone than any other condition [t(27)=9.54, p<.0001], with the large gapsize being more error prone than the small gap size [t(27)=-4.1, p<.0005].You may note that the occluded-truncated trials did not differ fromthe whole trials (See Table 1 for raw data) for the small or large gapsizes in terms of reaction times [t(27)=-.05 for small and t(27)=.399 forlarge] or error rates [t(27)=-.19 for small and t(27)=.449 for large] (SeeTable 1). This was expected for the small gap size but seemscounterintuitive for the large gap size, given the large effect found forthe occluded condition. This finding probably has to do with the choice ofstimulus configurations. Essentially, the occluded-t trails are present asa result of the need for notched trials and the desire to have all possiblepairings of items presented throughout the experiment. Thus, the largegap size for these trials refers to the gap in the truncated occluder andnot to the amount of overlap between the two items. This result, then,Early Completion24implies that the important variable for these stimuli is the amount ofoverlap and not the notch size in the occluding distractor.In comparing same and different trials a somewhat surprising speed-accuracy trade off was discovered such that same trials were faster thandifferent trials [E(1,9)=31.7, MS e =4069.5, p<.0003] although more errorprone [F(1,9)=5.8, MS e=.002, p<.05]. This trade off was entirely due to thealready mentioned high error rates in the notch condition for both thesmall and large gap sizes.Discussion For small amounts of overlap, subjects were able to respond to theoccluded objects in the same way as to the whole objects. Both of theseconditions were significantly faster than the notched condition. Thus,according to both of our operational definitions subjects seem to be usingthe early completion process for these small gap stimuli. This replicatesthe Gerbino and Salmaso (1987) finding.For large amounts of overlap, however, the occluded trials tooksignificantly longer than whole trials. The notched condition was alwaysslower and more error prone than either of the occluded or wholeconditions. In terms of our operational definitions these results aresomewhat ambiguous. Since the occluded trials were faster than theEarly Completion25notched trials we must assume that some information from early in thevisual stream is aiding the later categorization of the object fragment.However, since the occluded condition was slower than the wholecondition there must be some degree of failure of the early completionprocess.In terms of the two goals, the original study showing that the earlyvision strategy can operate on slightly occluded objects was replicated.As well, for large amounts of overlap, occluded objects required a longeramount of time and were more prone to errors than whole objects. Thus,the operation of this early strategy does indeed seem to be spatiallylimited in its extent of operation.Experiment 2aThe purpose of the Experiment 2 was to investigate the spatial limitof early completion in a more systematic way. In Experiment 1 therewere only two gap sizes. As well, the information obtained from the largegap condition was equivocal in terms of the two operational definitions.For small gap sizes, completion operated at full efficiency. The other endof the continuum where completion fails entirely was not found. Instead,Early Completion26for the large gap sizes, there was some evidence of completion by oneoperational definition but not by the second.There were two methodological simplifications from Experiment onewhich were hypothesised to reduce the overall number of processingstages that subjects must engage. First, the stimulus set size wasdecreased from three items to two, thus reducing the memory loadrequired to recognize the shapes. Second, the task was changed frommatching to a two alternative forced choice identification. This had theadvantage of eliminating the need for visual comparisons. The shapes usedwere circles and squares which could be either black or white on aneutral gray background. The subjects' task was to identify the blackobject as either a circle or a square. The white item was alwaysirrelevant and to be ignored by the subjects. By reducing the task demandson the subjects, the added variability from other processing stages wouldbe presumably reduced.The gap size measurements are given as the percentage of area thatwas covered by the occluding object. The four levels tested in thisexperiment were 25%, 50%, 70%, and 85%.The conditions tested were also slightly altered in this experiment totest several different hypotheses. The occluded-truncated condition wasEarly Completion27replaced with a whole-truncated condition. The whole-truncated conditionwas added to see if manipulations of the occluded, irrelevant distractorhad any effect on identification of the whole target. The occluded-truncated condition was removed to keep the total number of conditionsdown. Again there were two control conditions, single and double whichhad, respectively, one or two non-overlapping items. As well, there was anotched-alone condition which was included to see if the effect of adistractor varied with the size of the gap.Method Subjects. There were four right handed males and six right handedfemales. All subjects had normal or corrected to normal vision. Subjectswere drawn from the same subject pool as Experiment 1.Stimuli. A display consisted of a stimulus set displayed 6° to theright or left of fixation^Stimuli consisted of circles (2.2°) and squares(2.0°) which could be either black or white. The background was a neutralgray where 50% of the pixels were black and 50% were white (see figure2 right hand side).There were twenty two conditions. Twenty of these corresponded to afour by five factorial design where level of occlusion and trial type werethe two factors. The levels of occlusion that were tested included 25, 50,Early Completion2870, 85 percent occluded objects. The following trial types were included:whole, occluded, notched, notched-alone, and whole-truncated. The lasttwo conditions, which were not represented in the factorial design, werethe control conditions single and double. Note that the notched-alonecondition was identical to the notched condition except that no distractorwas presented behind the fragment.The black item always appeared on the midline of the screen. If thetarget was on the right side of the screen the distracting object couldappear above and to the right or below and to the left of the target. If thetarget appeared on the left side of the screen, the distracting objectcould appear above and to the left or below and to the right of the target.The side of presentation and the location of the distractor were randomlychosen from trial to trial by the computer.Apparatus. The equipment and software used were the same as inExperiment 1.Procedure. Overall procedures, display durations, and feedback wereidentical to Experiment 1.The task was to indicate the identity of the black item. Subjects usedthe index finger of one hand on the keyboard to indicate circle and theEarly Completion29index finger of the other hand to indicate square. The side of respondingwas counterbalanced across subjects. There were 20 blocks of 50 trials.ResultsAfter the first block of trials and the first three trials of eachsubsequent block were removed there were 893 trials to analyze persubject. Reaction times were submitted to a four by five repeatedmeasures factorial ANOVA. The first factor was level of occlusion andincluded 25%, 50%, 70%, and 85% occlusion. The second factor wascondition and included whole, occluded, notched, notched-alone, whole-truncated. Both main effects were significant as was the interaction ofthe two [E(4,36)=37.5, MS e=880.5, p<.0001 for type of trial; E(3,27)=31.6,MS e=536.6, p<.0001 for level of occlusion; F(12,108)=6.2, MS e=281.3,p<.0001 for the interaction] (See Figure 7 and Table 2).It had been predicted that for small amounts of overlap the occludedcondition would be as fast as the whole condition and much faster thanthe notched condition. For large gap sizes the reverse was predicted suchthat the occluded condition would be much slower than the wholecondition and not differ from the notched condition. Contrary to our firstprediction, the occluded trials were significantly different from theEarly Completion30Figure 7. Results of Experiment 2aEarly Completion31whole trials at all levels of occlusion including the smallest [t(108)=3.4,p<.0014 for the 25% occlusion condition]. In fact the occluded conditionwas not different from the notched condition overall [t(36)=.614]. Only atthe smallest level of occlusion (25%) was the occluded condition fasterthan the notched condition [t(108)=2.4, p<.02] as was predicted.Thus we confirmed three of our four predictions. First, for small gapsizes, the occluded condition was faster than the notched condition.Second and third, for large gap sizes, the occluded condition was equal tothe notched condition, and much slower than the whole condition. Theprediction that for small gap sizes the occluded condition would be asfast as the whole condition was not confirmed.Discussion For large gap sizes the occluded condition required the same amountof time as the notched condition. Both of these conditions were slowerthan the whole condition. For the smallest gap size (25%) it was foundthat the occluded condition was faster than the notched condition but notas fast as the whole condition. Thus the 25% level of occlusion wassimilar in pattern of results to the large gap size condition inExperiment 1.Early Completion32This experiment has isolated the opposite end of the continuum fromthe first experiment, that is, the end where there is no informationavailable to later cognitive levels of processing from the earlycompletion process. Experiment 2b was run to sample the levels ofocclusion between 3% and 46%.Experiment 2bThis experiment was conducted to investigate at what point thecompletion mechanism fails. The previous experiment showed that by 50%occlusion there is no evidence of completion and that at 25% there wasmarginal evidence for completion. The levels sampled for this experimentfell between 3% and 46% to give a more fine grained picture of thisimportant region. From the previous experiment we predicted that forocclusion levels greater than 25% there should be no completion while at25% there should be some completion. At some point below the 25% levelthere should be evidence for completion by both of the operationaldefinitions discussed previously.Method Subjects. Six female and four male right handed subjects were drawnfrom the same subject pool as in previous experiments.Early Completion33Stimuli. There were twenty-eight stimulus configurations. Twenty-five of these corresponded to a five by five factorial design where thetwo factors were condition (whole, occluded, notched, occluded-t, whole-t) and level of occlusion (3%, 5%, 13%, 25%, 46%). The remaining threeconditions were the control conditions, single, double, and touch. In thetouch condition the target and distractor did not overlap but were indirect contact. All other details were the same as in Experiment 2a.Apparatus. All details were the same as in previous experiments.Procedure. There were several changes from the previous experiment.A fixation point was added for 332 ms before the target presentation andduring the target presentation to further reduce eye movements. As well,the number of blocks and trials was changed to 15 blocks of 60 trials.Results After removing the first block and first three trials of eachsubsequent block there were 798 trials per subject to be analyzed. Eachsubjects mean reaction time was submitted to a five by five repeatedmeasures ANOVA where the two factors were level of occlusion (3, 5, 13,25, 46 percent occlusion) and condition (whole, occluded, notched,occluded-t, whole-t). Both main effects and the interaction weresignificant [F(4,36)=6.6, MS e =704.7, p<.0005 for level of occlusion;Early Completion34F(4,36)=13.4, MS e=946.5, p<.0001 for condition; F(16.144)=4.1, MS e=496.6,p<.0001 for the interaction of the two effects] (See Figure 8 and Table 3).In testing the a priori predictions regarding the small gap sizes itwas found that for 3% occlusion the whole, occluded and notchedconditions did not differ [t(144)=-1.7 for occluded vs. whole; t(144) =-1.6for occluded vs. notched; t(144) =0.2 for whole vs. notched]. Thus subjectsdid not find the three percent notch disruptive in any condition. At fivepercent the notched condition was slower than the whole condition[t(144)=-2.4, p<.05] but not than the occluded condition [t(144)=-1.3]. Thewhole and occluded conditions did not differ [t(144)=1.1]. At thirteenpercent the occluded and whole conditions did not differ [t(144)=1.5] andboth were faster than the notched condition [t(144)=-4.4, p<.0001 forwhole; t(144)=-2.9, p<.005 for occluded]. This is the first indication thatcompletion is occurring. At twenty-five percent the notched conditionwas slower than the occluded condition [t(144)=-2.6, p<.05] and the wholecondition [t(144)=-5.0, p<.0001]. The occluded condition was slower thanthe whole condition [t(144)=2.4, p<.05]. Once again, we have someevidence for completion. At 46% occlusion the notched condition wasslower than both the occluded condition and the whole condition [t(144).-8.6, p<.0001 for whole vs. notched; t(144)=6.2, p<.0001 for occluded vs.A■••.^I^I6H:t— occluded—0— whole.•—' notchedEarly Completion351,7680660 -E640 -0Ei:•co.«.00ere620 -AA •600 - /•• 0580 I . I0^10^20^30 40^50Percent OccludedFigure 8. Results of Experiment 2bEarly Completion36notched]. The occluded condition was slower than the whole condition[t(144)=-2.3, p<.05]. Note, however, that the difference between theoccluded and whole conditions for the 46% occlusion (62ms) is muchgreater than at 25% occlusion (24ms). This difference between theoccluded condition at 46% and 25% occlusion is significant [t(144)=-3.3,p<.01].The control conditions occluded-t and whole-t did not differ fromtheir respective experimental conditions for any level of occlusion[t(144)<1.0 for all comparisons].The error analysis revealed no significant main effects and nointeraction [E(4,36)=1.4, MS e =.001 for level of occlusion; E(4,36)=0.8,MS e =.001for condition; F(16.144)=0.8, MS e=.001 for the interaction of thetwo effects].Discussion It was predicted that highly occluded objects would not be completedquickly while partly occluded objects would be. In Experiment 2a it wasshown that highly occluded objects are not completed while marginallyoccluded objects (25%) were partly completed. In Experiment 2b it wasshown that partly occluded objects are completed and marginallyoccluded objects are again partly completed. There seems to be anEarly Completion37important break point, for these stimuli, at the 25% occlusion level suchthat below this level there is clear evidence for the operation of earlycompletion while above this level there is clear evidence for failure ofcompletion. At 25% the data is equivocal. By the first operationaldefinition for completion there is a failure while by the secondoperational definition completion succeeds. It is thus concluded thatsome information is passed forward to later vision processes in thisintermediate condition.It is interesting to note that overall reaction times in Experiment 2bare longer than those in Experiment 2a (Compare Table 2 and Table 3).This 110ms effect did not seem to change the pattern of results and hadminimal effects on the difference between the single and double controlconditions (29ms in 2a and 26ms in 2b). This effect can be attributed tothe addition of the fixation point before and during the presentation ofthe stimuli. This central fixation point may have had two related effects.First, overall number of eye movements may have been reduced by givingsubjects something on which to fixate. Second, some attention may havebeen diverted from the task to maintain this fixation which would have aneffect on the identification process but should not affect the pre-attentive completion process. This may be taken as suggestive evidenceEarly Completion38that manipulations of later attentional processes do not affectcompletion as was proposed in the introduction.Having shown that the early completion processes is spatiallylimited and further having delineated the extent of this limitation, thenext experiments were conducted to show that this result is not taskdependent.Experiment 3It has been proposed that partly occluded objects are completed earlyin the visual stream and further that this completion process is spatially(Experiment 1 and 2) and temporally (Sekular & Palmer, 1992) limited.The next two experiments explore how separable this mechanism is fromother visual routines. The current experiment investigates effects ofchanging the subjects' task. If completion occurs early in the visualstream as has been proposed, then we should expect the same spatiallimitations regardless of task. As well, in the two previous tasks(matching and identification) it was to the subjects' advantage tocomplete the object. That is to say, reaction times would decrease to theextent that subjects completed the target. The current task, decidingwhether an object was retinally complete or incomplete, requires thatEarly Completion39the subject not complete partly occluded objects. To the extent that thecompletion processes is automatic subjects' reaction times in theoccluded condition should increase when completion occurs. This test ofautomaticity is similar to that used by Enns and Rensink (in press) in thesearch experiment discussed earlier.Thus, there are two purposes for the current experiment: todetermine whether the spatial limitations proposed for completion aretask independent of other response relevant processes, and to furtherinvestigate the automatic nature of completion. The operationaldefinitions of completion must be redefined for this new task. There areagain two definitions. First, completion will be assumed when theoccluded condition is slower than the notched condition. In these twoconditions the retinal stimulation of the two fragments is identical as isthe required response. The only difference between these conditions isthe extent to which completion can be accomplished (i.e. completionoccurs in the occluded condition and not in the notched condition). Thesecond definition states that completion will be assumed when theoccluded condition is slower than the whole condition. For these twoconditions, when completion occurs, the internal representation issimilar, but the response required is differentEarly Completion40It was predicted that reaction times for the notched and wholeconditions would not differ and should be relatively unaffected by thelevel of occlusion. To reiterate, the occluded condition should takesignificantly longer than the whole and notched conditions for small gapsizes but not for large gap sizes. In addition, the point at whichcompletion begins to fail should occur at approximately the same level ofocclusion as in the previous experiment (i.e. 25% occlusion).Method Subjects. Four right handed males and six right handed femalesparticipated for course credit. Subjects were drawn from the samesubject pool as for previous experiments.Stimuli. The item configurations were identical to Experiment 2awith the exception that the levels of occlusion sampled included 5, 13,25, 36, 54 percent occlusion. As a reminder, the conditions included werewhole, occluded, notched, occluded-t, whole-t, single, and double.Apparatus. The equipment and software used were the same asprevious experiments.Procedure. Overall procedures, display durations, and feedback wereidentical to Experiment 2b.Early Completion41The task was to indicate whether the black item was complete orincomplete. Subjects used the index finger of one hand on the keyboard toindicate complete (whole) and the index finger of the other hand toindicate incomplete (notched). The side of responding wascounterbalanced across subjects. It was emphasised that the items in theoccluded condition were retinally incomplete and thus required the sameresponse as the notched conditions. There were 15 blocks of 60 trialseach.ResultsAfter the first block of trials and the first three blocks of eachsubsequent block were removed there were 798 trials for each subject.The mean reaction times for all correct responses were submitted to fiveby five repeated measures ANOVA where the two factors were level ofocclusion (5, 13, 25, 36, 54 percent occlusion) and condition (whole,occluded, notched, occluded-t, whole-t). Both main effects weresignificant as was the interaction of the two [E(4,36)=36.6, MS e=875.0,p<.0001 for level of occlusion; F(4,36)=6.4, MS e=4377.8, p<.0005 forcondition; E(16,144)=506, MS e=847.8, p<.0001 for the interaction of thetwo effects].Early Completion42Figure 9. Results of experiment 3Early Completion43In testing the a priori predictions it was found that the occludedcondition was significantly slower than the notched condition for alllevels of occlusion except the largest [t(144) =6.5 for 5%; t(144)=4.1 for13%, t(144)=3.7 for 25%; t(144) =2.7 for 36%, p<.0005 for allcomparisons; t(144)=1.53 for 54%]. This indicates that the occludedobject was completed for all levels of occlusion except for the 54%. Theoccluded condition was also significantly slower than the whole conditionfor the first three levels of occlusion [t(144)=8.2 for 5%; t(144)=3.4 for13%; t(144)=3.6 for 25%, p <.0005 for all comparisons; t(144)= 1.1 for36%; t(144)=0.3 for 54%]. The whole and notched conditions did not differfrom each other for any level of occlusion [t(144)<1.9 for all levels] (Seefigure 9 and Table 4).The control conditions occluded-t and whole-t did not differ fromtheir respective experimental conditions except at the 25% occlusionlevel where the occluded-t was faster than the occluded condition[t(144)=2.2, p<.05]. As well, at this level the occluded-t condition did notdiffer from the notched [t(144)=1.5], whole [t(144)=1.3], or whole-t[(144)=1.0] conditions indicating that for this control condition nocompletion was occurring.Early Completion44The error analysis indicated two significant main effects and asignificant interaction [F(4,36)=9.7, MS e =.001, p<.0001 for level ofocclusion; E(4,36)=6.0, MS e =.002, p<.0009 for condition; .E(16,144)=4.7,MS e =.001, p<.0001 for the interaction of the two effects]. The interactionwas due solely to the high error rate in the occluded and occluded-tconditions for the 5% occlusion level.Discussion As was predicted, completion occurred for small gap sizes (<36%) butnot for large gap sizes (54%). This shows that the spatial limitationsproposed for the completion mechanism are independent of task demands.Although there was evidence for completion at higher levels of occlusion,there are signs of difficulty for completion at the 25% level when lookingat the occluded-t condition.The similarity of results between Experiment 2 and 3 support thenotion that completion is a separate process from that used to identify orcatagorize these target items. Separable does not imply fullindependence. At present we do not have a precise enough definition ofthe involved processes to investigate the independence or interaction ofthese processes. The separable nature of the proposed object completionis still tentative. To strengthen the argument for separability the fourthEarly Completion45experiment was run using both static and moving objects to test ifmotion would affect completion while leaving other processes unaltered.Experiment 4There are several experiments in the literature that haveinvestigated the interaction of completion and movement. Kellman &Shipley (1992) have drawn a tentative link between motion inducedsubjective contours and static subjective contours and have furtherimplied a single mechanism operating in these two situations. As well,Gibson and collogues have shown that depth segregation can beaccomplished within displays where the only cue to an occlusion contouris motion of random dots (Gibson, Kaplan, Reynolds, & Wheeler, 1969;Kaplan, 1969).More recently Shimojo, Silverman & Nakayama (1989) have shownthat completion occurs more readily in moving displays than in staticdisplays. The added information from the accretion and deletion ofsurface properties at the occlusion boundary must aid in the segregationof the two objects into separate depth planes which should aid in thetask of completion.I will be using motion as a tool to show that early completion isseparable from later categorization and response based processes. It isEarly Completion46predicted that completion will be more effective on moving targets due tothe added information provided at the occlusion boundary. At the sametime there should be no difference between the moving and static targetsin the whole condition. Differences may surface in the notched conditiondue to the production of illusory contours induced by the motion of thetarget behind an invisible occluder.The important comparisons to be made for the purposes outlinedabove are between the static and moving target for the occludedcondition. As well, a set of conditions was added where all of theelements in the display move in concert with the target. This display hasall the same information as the moving target conditions without theaccretion and deletion at the occlusion boundary. For this reason theconditions in which all of the items move in unison may serve as bettercontrols for the target move conditions than the static conditions.Method Subjects. Fourteen subjects were selected from the same subjectpool used previously. Two subjects were discarded: the first failed tocomplete all of the blocks, and the second had error rates above 25% inthe second session. This left nine right handed females and three righthanded males. All subjects had normal or corrected to normal vision.Early Completion47Figure 10. Sample Stimuli used inExperiment 4Early Completion48Stimuli.^The item configurations were identical to Experiment 3with exception that the levels of occlusion sampled included 5, 20, 28,39, 54 percent occlusion. The conditions included were whole, occluded,notched, occluded-t, whole-t, and double. There was an added factorof motion which had three levels including static, all move, and targetmove. The static displays were a replication of the previous experiment.In the all move conditions every visible contour of the target anddistracting item moved in concert (See figure 10). Thus there was noaccreation and deletion of information at the occludion boundary. In thetarget move conditions only the target moved and all other objects,visible or not were stable. Thus, there was accreation and deletion ofinformation at the occlusion boundary even in the notched condition wherewhere there was no visible occluder. This condition in particular wasperceptually odd because of the compelling percept of a complete itemmoving behind a "phantom" occluder. The motion was in a directionperpendicular to the line on which the distractor could be presented(remember that the distractor could be presented either to the right andabove or left and below a right target and to the left and above or rightand below of a left target). Thus a right target moved along the diagonal45° to the right. A left target moved along the diagonal 45° to the left.Early Completion49The items were presented, sequentially, in three positions on each trial togive the impression of motion. The percent occlusion remained constantthrough all three of the target positions. The target oscillated at a rateof 2.8° per second and moved over a total distance of 0.2°. The stimuliwere presented for a total of 265 ms. The added 65ms in relation toprevious experiments was necessary due to machine limitations and thedesire to have all three positions represented equally.Apparatus. All equipment and software was the same as in previousexperiments.Procedure.^Overall procedure was the same as in previousexperiments with the exception that subjects ran in two session eachconsisting of 20 blocks of 50 trials. The added session and blocks werenecessary due to the added conditions.ResultsThe first block of both sessions and the first three trials of all otherblocks were discarded as practice. This left 1786 trials per subject to beanalyzed. The error free reaction times were submitted to a three factorrepeated measures ANOVA where the factors were motion (static, allmove, target move), condition (whole, occluded, notched, occluded-t,whole-t), and percent occlusion (5%, 19%, 28%, 39%, 54%) . The mainEarly Completion50effects of condition and occlusion were significant while that for motionwas not [F(4,44)=5.2, MS e =9313.8, p<.002 for condition; F(4,44)=47.1,MS e =1317.1, p<.0001 for occlusion; E(2,22)=0.1, MS e =730.6 for motion].These effects were moderated by two higher order interactions. The firstbetween motion and condition [F(8,88)=2.4, MS e=909.4, p<.05] and thesecond between condition and occlusion [E(16,176)=7.3, MS e=924.2,p<.0001]. The three way interaction between motion, condition andocclusion was not significant [.E(32,352)=1.2, MS e=824.8].The motion by condition interaction was due primarily to differencesbetween the target move and the all move conditions for the occludedtrials and the whole-t trials. For the occluded conditions, target movewas slower than all move [t(96)=-2.2, p<.05] while for the whole-tconditions, target move was faster than the all move [t(96)=2.6, p<.05].The differences in the whole condition and the occluded-t conditions werein the same direction but not significantly different. Thus collapsingacross levels of occlusion subjects found the occluded conditions moredifficult when the target only moved as opposed to when all the itemsmoved. The static conditions were intermediate to the other twoconditions and not different from either. This thus confirms our a prioriEarly Completion51predictions that the occluded object would be best completed in thetarget move conditions.When considering the occluded condition in more detail by examiningacross the five levels of occlusion, it was found that only for the 20%occlusion level was the target move case slower than the other twomotion cases R(352)=-2.3, p<.05 for the static; t(352)=-3.3, p<.01 for theall move]. This again confirms the prediction that completion is moreefficient in the target move case for this level of occlusion (see Figure10 and Tables 5, 6 and 7).The errors from each condition were submitted to the same ANOVA.Again the main effects of condition and occlusion were significant[E(4,48)=3.8, MSe=15.9, p<.01 for condition; .E(4,48)=4.5, MS e=12.5, p<.01]for occlusion].^As well the interaction of condition and occlusion wassignificant [E(16,176)=2.9 MS e =9.6, p<.001].^In general the errorsfollowed the same trends as the reaction times and were not furtheranalyzed.Practice effects were analyzed by splitting each of the two sessionsin half and submitting the resulting four blocks of data to a four factorrepeated measures ANOVA where the four factors were practice, motion,condition, and occlusion. There was an overall effect of practiceEarly Completion52[F(3,33)=6.2, MS e =71828.0, p<.002].^As well, practice interacted withcondition [E(12,132)=4.4, MS e =7944.0, p<.0001] and with occlusion[F(12,132)=12.7, MS e =5308.7, p<.0001]. As well, there was a three wayinteraction between practice, condition and occlusion [.E(48,528)=2.6,MS e =4070.7, p<.0001]. These higher order interactions were due to anunexpected reversal for the 28% and 39% occlusion such that early in bothsessions subjects found the notched and whole conditions difficult andthe occluded condition easy. In the latter half of both session the notchedand whole conditions became much easier while the occluded conditionremained at about the same level. This uncharacteristic reversal was notpresent in any of the past experiments and is thus attributed to theintroduction of motion. At present the only explanation that can beconjectured is one in which subjects are relying on later processes forthese boundary conditions where completion is beginning to fail and thatthese later cognitive processes do not follow the rules proposed for theearly completion process. Note that by 54% occlusion the results arefairly stable and in the predicted direction so that the difficulty ininterpretation lies only at the 28% and 39% occlusion where subjects'may be using a mix of strategies, both within a given trial and acrosstrials, based on the amount of information available from early processes.Static— 0— All in Motion- I=1- Target in motionWhole conditionsEarly Completion53Occluded Condition For All Motion Cases600 -580 -IiiE560 -dI.-• 540 -O^coecc^520 -500 -480 I^1^1^1^1 0^10 20 30 40 50^60Percent OccludedFigure 11. Results from experiment 4Early Completion54Discussion It had been predicted that the introduction of motion for the targetonly would enhance the completion process and thus make occluded trialmore difficult for these trials when compared to the all move conditionsand to the static conditions. This was confirmed. The effect wasprimarily limited to the 20% occlusion level. This further supports theproposal that early completion is separable from later decisionprocesses. This separability should not be considered a strong form ofindependence as there was some interaction between early and lateprocesses across practice. The exact nature of the practice effects awaitfurther experimentation.Early Completion55General DiscussionThe question posed at the beginning of this study dealt with whetheror not all objects are completed and what spatial factors influence thefailure of completion. Sekular and Palmer (1992) showed that completionfailed when there was insufficient time allowed for foveated figures.Experiment 1 showed that completion fails for non-foveated targetspresented for 200ms with large gap sizes as opposed to small gap sizeswhich were completed. Experiment 2 expanded on these results to showthat at about 25% occlusion there is only partial completion of theoccluded object. Beyond this level there was a complete failure ofcompletion. Experiment 3 replicated this failure of completion using adifferent task (categorization as complete or incomplete as opposed to anidentification task). The task irrelevant nature of the completion failureis taken as support for the separability of early completion from laterresponse dependent processes. Experiment 4 demonstrated that earlycompletion is enhanced by placing the target in motion relative to thebackground items while leaving later discrimination processes, for themost part, unaffected.Overall, completion can be characterized as a rapid, automatic, andspatially parallel process that begins to break down when large gap sizesEarly Completion56are presented for short durations. In the real world this implies thatearly vision attempts to complete partly occluded objects and is, for themost part, successful. Difficulties in completion arise when the objectto be completed is highly occluded or only briefly visible. In these caseswhere there is a failure of completion, the observer must either delayresponding or direct attention to aid in the interpolation of the retinalfragments. The exact role that attention plays in completion was notaddressed in this paper. However, a comparison of Experiments 2a and 2bseems to indicates that completion is independent of attentionaldemands. This begs further experimentation to ascertain the benefits thatattention could lend to completion.Current theories of visual information processing that utilize theearly-late dichotomy have typically thought of the first stage asregistering only simple 2-D features as they appear on the retina. Thelatter stages are thought to integrate these free floating features bydirecting attention to a given location in space. (Treisman, 1988;Treisman & Gelade, 1980; Treisman & Sato, 1990). According to FeatureIntegration Theory (Treisman & Gelade, 1980), the interobject relation ofocclusion should not be computed early in vision since this attributerequires the computation of spatial relations between objects which areEarly Completion57themselves not registered until later processing stages. Thus, thedemonstration that occluded objects are completed preattentively (Enns &Rensink, in press) counters the simplistic view of early vision. Instead,early vision should be conceptualized as a hard wired system thatcomputes scene-based 3D attributes that in the past have been shown tobe relevant (Enns, 1992). I am argueing here not for a simple addition ofnew 3D features to the already present framework of 2D features (Enns,1990) but for a reconceptualization of what are the primitive elements ofperception (Enns & Rensink, 1993). Support for this line of argumentcomes not only from studies of occlusion (Enns, 1992; Enns & Rensink, inpress), but also from search studies showing preattentive registration ofshape from shading, direction of lighting and the recovery of slant fromtexture (Aks, 1993; Aks & Enns, 1992; Enns, 1990; Enns & Rensink, 1992;Enns & Rensink, 1993; Kleffner & Ramachandran, 1992). These 3Dattributes of the scene can not be derived from the mere registration of2D features but instead require the computation or recognition of certaincomplex invariant attributes in the environment. The fact that earlyvision is sensitive to these attributes speaks against the simplistic viewpresented in Feature Integration Theory.The current work follows from the finding that occlusion relationsEarly Completion58are registered preattentively. For an attribute of the image or the sceneto be computed in parallel across the visual scene, there must be someoperator that acts at every location of the image over a finite spatialarea, called a neighbourhood of operation (Enns, 1992; Enns & Rensink,1993; Zucker, 1987). The assumption that completion is spatially limitedwas tested and found to be true, adding further support to the notion thatocclusion relations are registered early within finite neighbourhoods ofoperation.Preattentive processes have for the most part been characterized asautomatic data-driven processes that are beyond the subjects voluntarycontrol. Experiments 3 and 4 speak to this issue and support theconjecture that completion is indeed automatic. Recall that in thoseexperiments it was to the subjects advantage not to complete the object.Despite this, subjects completed the occluded figure, thus slowingperformance on the task for small gap sizes. Note, however, that theautomatic nature of completion does not prevent it being manipulated byscene attributes (i.e. the kinetic nature of the objects in Experiment 4).This last experiment supports the notion that completion is an automaticprocess that is distinct in terms of operating stages from later attention-based processing stages. These later stages are presumably used toEarly Completion59identify or describe the objects passed down the stream by the earlyvision processes.As this is still a new area of study, there remain many questionswhich need to be answered in an objective, performance-based way. Theexact relation between alignments of attention, both voluntary andinvoluntary, and completion need to be further delineated. Specifically,when attention is aligned with an area of occlusion, will completion beimproved? To address this question one could use a Posner (1980) cueingparadigm in conjunction with the present notched task to discern if onvalid trials completion is more efficient than on neutral or invalid trials.An additional area of study that requires further research is theexact time course of the completion process. Sekular and Palmer (1992)showed that partly occluded objects are completed by 200ms but theyused an indirect priming paradigm and required that a third object bepresent in the probe to eliminate apparent motion affects. As well, theyonly used one level of occlusion. Are the more highly occluded objects notcompleted at all, as implied earlier, or do they simply take more time tocomplete? This question could be addressed by conducting a microgenesisexperiment similar to the earlier work (Sekular & Palmer, 1992) andusing primes of different levels of occlusion.Early Completion60The finding that the early completion process can be manipulatedindependently of the later, presumably attentional, process begs thequestion: Can the reverse be done? Can the attentional process bemanipulated while leaving early completion unaffected? To address thisquestion, one could manipulate the complexity of items to be identified ormatched while leaving the amount of occlusion unaltered. It is predictedthat the effect of complexity would be orthogonal to the effect of amountof overlap. This would further speak to the question of the role thatattention plays in completion.The evidence presented in the current paper is consistant with a localcompletion process. That is, one in which completion is based on theattributes present at the occlusion boundary (Kellman & Shipely, 1992).An alternate view, where completion is affected by global attributes ofthe figure to be completed, has been recently presented by A. Sekular(submitted). According to this theory, there are two factors contributingto the completion of occluded objects: a top-down influence based onobject symmetry and a bottom-up input based on local junctioninformation. This hybrid model has a certain intuitive appeal but atpresent needs further refinement. Evidence for the global aspect ofcompletion comes from high speed priming studies where the degree ofEarly Completion61symmetry and the axis of symmetry were manipulated. It was found thatfor fourfold symmetrical objects completion was entirely predicted byglobal theories (Sekular & Gaber, 1993). When the symmetry was onlytwofold or onefold, the axis of symmetry determined the mix of globaland local contributions with local global theories dominating when thesymmetry was across the vertical or oblique axis but not the horizontalaxis. Due to the indirect nature of priming paradigms it would beappropriate to conduct further research into the exact contribution ofglobal and local factors into the completion process. An additional notefor this area of research is the exact connotations from the terms globaland local which are in dire need of stringent definition.One final area which requires investigation is the role that depthsegregation or figure-ground relations plays in completion. To individuateand recognize an object, two things must occur: the figure must beseparated from the ground and the figure must then be completed orinterpolated. The ordering of these two processes is a question that mustbe answered. The current study makes no mention of depth or segregationin depth. This is because the work presented does not address thisquestion, although it is intimately related.In closing, it seems that there is still a great deal of work to be doneEarly Completion62before we can understand completion and its role in everyday perception.There is strong support now that completion does occur early in visualprocessing and in parallel across the scene. Further attributes of thisprocess and the delineation of other potential completion processesawaits further work.Early Completion63ReferencesAks, D.J. (1993). The Analysis of Slant-From-Texture in Early Vision.Unpublished doctoral dissertation, University of British Columbia,British Columbia, Canada.Aks, D.J., & Enns, J.T. (1992). Visual Search for Direction of Shading IsInfluenced by Apparent Depth. Perception & Psychophysics,  52, 63-74.Enns, J.T. (1990). Three-dimensional features that pop out in visualsearch. In D. Brogan (Ed.) Visual Search. Taylor & Francis: London, pp.37-45.Enns, J.T. (1992) The Nature of Selectivity in Early Human Vision. In B.Burns (Ed.), Percepts. concepts. and categories: The representation and processing of information (pp. 2-34). Amsterdam: ElsevierScience Publications.Enns, J.T., & Rensink, R.A. (1993). A model for the rapid interpretation ofline drawings in early vision. In D. Brogan, A. Gale, K. Carr (Eds.)Visual Search 2. London: Taylor & Francis, pp.73-89.Enns, J.T., & Rensink, R.A. (in press). An Object Completion Process inEarly Vision, To appear in A.G. Gale (Ed.), Visual Search III: Proceedings of the Third International Conference on Visual Search.London: Taylor & Francis.Gerbino, W., & Salmaso, D. (1987). The Effect of Amodal Completion onVisual Matching, Acta Psychologica,  65, pp. 25-46.Gibson, J.J., Kaplan, G.A., Reynolds, H.N., Jr., & Wheeler, K. (1969). Thechange from visible to invisible: A study of optical transitions.Perception & Psychophysics, 5, 113-116.Greenhouse, S.W., & Geisser, S. (1959). On methods in the analysis ofprofile data,  Psychometrika,  24, 95-112.Early Completion64Huynh, H., & Feldy, L., S. (1973). Conditions under which mean squareratios in repeated measurement designs have exact F-distributions.Journal of american Statistical Association, 65, 1582-1589.Kanizsa, G., & Gerbino, W. (1982). Amodal Completion: Seeing or Thinking?.In J. Beck (Ed.) Organization and Representation in Perception.Hillsdale, NJ: Erlbaum.Kaplan, G.A. (1969). Kinetic disruption of optical texture: The perceptionof depth at an edge, Perception & Psychophysics,  6, 193-198.Kellman, P.J., & Shipley, T.F. (1992). A Theory of Visual Interpolation inObject Perception, Cognitive Psychology, 23, 141-221.Kleffner, D.A., & Ramachandran, V.S. (1992). On the Perception of Shapefrom Shading, Perception &Psychophysics, 52, 18-36.Posner, M.I. (1980). Orienting of Attention, Quarterly Journal ofExperimental Psychology, 32, 3-25Ramachandran, V.S. (1992). Filling in Gaps in Perception: Part I, CurrentDirections in Psychological Science,  1, 199-205.Ramachandran, V.S. (1993). Filling in Gaps in Perception: Part II. Scotomasand Phantom Limbs, Current Directions in Psychological Science, 2,56Rensink, R.A., & Enns, J.T. (1992). VScope Software. micropsych software,U.B.C.Sekular, A. (submitted). Perceptual Completion of Partly occluded Objects:Evidence from the primed matching paradigm. Perception.Sekular, A.B., & Gaber, E. (July, 1993). Perceptual completion: A Local orGlobal Process? Paper presented at the Third Annual Meeting of theCanadian Society for Brain, Behaviour, and cognitive Science: Toronto,Ontario.Early Completion65Sekular, A.B., & Palmer, S.E. (1992). Perception of Partly Occluded Objects:A Microgenetic Analysis, Journal of Experimental Psychology: General, 121, 95-111.Shimojo, S., Silverman, G.H., & Nakayama, K. (1989). Occlusion and theSolution to the Aperture Problem for Motion. Vision Research, 29,619-626.Treisman, A. (1988). Features and Objects: The Fourteenth BartlettMemorial Lecture, The Quarterly Journal of Experimental Psychology,40A, 201-237.Treisman, A., & Gelade, G. (1980). A feature Integration Theory ofAttention. Cognitive Psychology, 12, 97-136.Treisman, A., & Sato, S. (1990). conjunction search Revisited, Journal ofExperimental Psychology: Human Perception and Performance, 16,459-478.Zucker, S.W. (1987). Early Vision. In S.C. Shapiro (Ed.), The encyclopedia ofartificial intelligence (pp. 1131-1152). NY: John Wiley.Early Completion66Appendix A - Raw Tabled DataTable 1. Results of Experiment 1: Raw Data (RT and Error) ^ 67Table 2. Results of Experiment 2a: Raw Data (RT and Error) ^ 68Table 3. Results of Experiment 2b: Raw Data (RT and Error) ^ 69Table 4. Results of Experiment 3: Raw Data (RT and Error) ^ 70Table 5. Results of Experiment 4 - Static Conditions:Raw Data (RT and Error) ^  71Table 6. Results of Experiment 4 - All Move Conditions:Raw Data (RT and Error) ^  72Table 7. Results of Experiment 4 - Target Only Move Conditions:Raw Data (RT and Error) ^  73Early completion67Table 1. Results of Experiment1: Raw DataReaction Times Gap Size Occlusion Conditions Small LargeWhole 733 (42) 722 (39)Occluded 748 (38) 853 (48)Occluded-truncated 744 (52) 713 (41)Notched 899 (58) 1006 (69)Different 938 (40) 971 (45)Type of ConditionControl^Conditions Single DoubleSameDifferent615721(27)(32)751949(51)(57)Error RatesGap SizeOcclusion Conditions Small LargeWhole 6.2 (2.0) 6.9 (3.1)Occluded 10.0 (4.5) 15.3 (2.7)Occluded-truncated 6.9 (2.9) 5.7 (1.6)Notched 23.2 (6.1) 42.3 (8.8)Different 7.2 (1.4) 9.0 (1.5)Type of ConditionControl Conditions^Single^DoubleSame 1.3 (0.7) 7.5 (2.0)Different .^5.3 (2.4) 8.4 (3.0)Early completion68Table 2. Results of Experiment 2a: Raw DataPercent OccludedConditions^25%^50%^70%^85%Whole 470 (18) 471Occluded 496^(19) 515Notched 514^(19) 515Notched-alone486 (21)465 (19)506484Whole-TWhole 4.2^(1.4) 1.1Occluded 2.7 (0.9) 3.6Notched 1.9^(0.9) 2.3Notch-alone 1.9^(0.8) 2.5Whole-T 2.5 (0.8) 2.0Reaction Time(17) 475 (18)(19) 540 (19)(20) 534 (20)(20) 512 (17)(21) 463 (20)Error Rates(0.6)^1.4 (0.5)(1.0)^4.0 (1.3)(1.2)^3.6 (1.6)(0.8)^2.8 (0.6)(1.0)^0.6 (0.4)486 (20)570 (23)573 (24)535 (28)486 (24)2.2 (1.7)8.2 (1.7)4.2 (1.1)3.4 (0.8)2.5 (1.1)The control condition had the following RTs and Error rates:Single: 455 (18), 1.5 (0.4)Double:484 (19), 2.0 (0.6)13%^25%^46%Reaction Time590 (20) 586605 (22) 610633 (23) 635602 (23) 601590 (20) 593Error Rate^(20 ^581 (23)^(20)^643 (25)(23)^667 (17)(21)^624 (19)^(24 ^592 (21)3.8 (1.0) 1.72.8 (1.5) 2.72.2 (0.8) 3.93.2 (1.0) 3.23.3 (1.1) 3.5(0.8)^3.1 (1.2)(1.1)^3.8 (1.4)(1.6)^4.7 (1.4)(1.1)^4.7 (1.1)2.9 (1.3)(1.6)Early completion69Table 3. Results of Experiment 2b: Raw DataPercent OccludedConditions 3% 5%Whole 607 (25) 588 (20)Occluded 590 (21) 598 (23)Notched 605 (21) 612 (21)Occluded-T 590 (21) 599 (22)Whole-T 603 (23) 589 (21)Whole 2.5 (0.9) 2.4 (0.7)Occluded 3.8 (1.7) 2.8 (1.3)Notched 2.1 (0.6) 3.7 (1.3)Occluded-T 0.8 (0.5) 3.2 (1.1)Whole-T 1.9 (1.3) 3.1 (1.0)Note the control condition rt and error rates were as followsSingle:Double:Touch:567 (18), 2.2 (0.9)593 (24), 0.6 (0.4)588 (22), 4.6 (1.1)25%^36%Reaction Time668 (41) 666 (37)713 (42) 680 (40)665 (35) 645 (43)685 (37) 673 (36)672 (38) 669 (34)Error Rate1.6 (0.7) 1.0 (0.5)2.0 (0.9) 1.0 (0.5)0.4 (0.4) 0.7 (0.5)2.9 (0.9) 1.8 (0.8)0.4 (0.4) 2.0 (1.3)54%670 (35)674 (40)654 (38)669 (38)661 (37)0.6 (0.4)0.7 (0.5)0.8 (0.5)0.0 (0.0)0.7 (0.7)Early Completion70Table 4. Results of Experiment 3: Raw DataPercent OccludedConditions^5% 13%Whole^675 (47) 683 (39)Occluded^782 (38) 727 (41)Notched^698 (43) 674 (41)Occluded-T 787 (38) 742 (39)Whole-T^686 (40) 671 (67)Whole^0.7 (0.4) 1.3 (0.7)Occluded^7.8 (2.7) 2.0 (1.2)Notched^1.6 (0.7) 1.1 (0.6)Occluded-T 12.7 (3.8) 4.6 (2.1)Whole-T^1.8 (0.6) 2.4 (1.1)Note the control condition rt and error rates were as followsSingle: 567 (18), 2.2 (0.9)Double: 593 (24), 0.6 (0.4)Touch: 588 (22), 4.6 (1.1)Early Completion71Table 5. Results of Experiment 4 - Static ConditionsCondition 5%whole 541 (26)occluded 598 (36)notched 527 (20)occluded-t 579 (25)whole-t 549 (22)whole 1.6 (0.8)occluded 4.1 (1.7)notched 1.3 (0.7)occluded-t 3.7 (1.6)whole-t^2.4 (0.8)Percent Occluded^521 ( 8)^534 (30)^552 (32)^531 (28)513 (26)^496 (28)551 (29)^534 (36)523 (16)^532 (22)(1.0) 1.9 (0.7)(0.7) 2.5 (1.8)(1.3) 1.0 (0.7)(0.9) 2.2 (1.0)(1.4) 3.9 (1.3)Error Rate2.31.22.81.62.520%^28%^39%^54%Reaction Time527 (22) 517 (15)528 (25) 515 (22)484 (27) 492 (27)508 (25) 507 (26)520 (17) 532 (22)2.9 (0.9) 0.6 (0.4)1.3 (0.7) 2.1 (1.0)0.0 (0.0) 1.3 (0.9)0.6 (0.6) 1.2 (0.6)0.9 (0.5) 1.9 (0.7)Reaction Time514 (12) 516 (21)513 (25) 524 (25)497 (30) 484 (22)507 (22) 506 (24)539 (22) 522 (18)2.1 (1.0) 2.7 (0.9)1.9 (1.4) 1.2 (0.7)1.7 (0.7) 1.2 (0.9)1.7 (1.0) 1.6 (0.9)1.4 (0.7) 0.6 (0.4)^(22)^520 (20)^(37)^537 (28)(30)^502 (30)(30)^517 (23)(12)^538 (20)Error Rate(0.8) 1.9 (0.7)(0.8) 2.2 (1.0)(1.0) 1.2 (0.6)(1.3) 2.8 (1.2)(0.6) 3.1 (1.0)Early Completion72Table 6. Results of Experiment 4 - All MoveConditionsPercent Occluded5%^20%^28%^39%^54%whole 5 9 (16) 534occluded 595 (27) 539notched 531 (27) 517occluded-t 596 (27) 550whole-t 544 (18) 540whole 2.2 (0.9) 1.5occluded 6.8 (1.7) 2.1notched 0.4 (0.4) 2.1occluded-t 2.3 (1.9) 2.1whole-t^2.8 (1.0) 1.8Early Completion73Table 7. Results of Experiemnt 4 - Target OnlyMove Conditions5%whole 526 (20)occluded 599 (27)notched 525 (26)occluded-t 588 (24)whole-t 520 (15)whole 1.9 (0.6)occluded 6.0 (2.0)notched 1.0 (0.7)occluded-t 4.5 (1.1)whole-t^0.6 (0.4)Percent Occluded20%^508 (18)^526 (18)578 (41)^532 (23)498 (27)^514 (24)570(27)^528 (23)540 (23)^526 (21)(0.7) 2.5 (1.3)(1.1) 3.4 (1.1)(0.7) 1.1 (1.1)(1.9) 2.3 (1.2)(1.0) 0.8 (0.5)Error Rate1.52.70.95.42.454%28%^39%Reaction Time506 (14) 519 (20)524 (26) 530 (26)488 (26) 488 (24)512 (28) 524 (33)509 (17) 515 (17)2.8 (0.7) 2.3 (1.2)1.7 (1.0) 2.1 (1.0)0.4 (0.4) 0.8 (0.6)0.4 (0.4) 1.2 (0.6)2.1 (0.8) 1.2 (0.6)

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0086297/manifest

Comment

Related Items