How do the internal details of the object contribute to

0 downloads 0 Views 268KB Size Report
be difficult from that viewpoint (for a detailed discussion see Biederman and Bar .... the left in depth from the foreshortened view (see figure 1). ..... TV set. Clarinet. Pan. Stapler. Violin. Desk. Penguin. Table. Watering can ... Sewing machine.
Perception, 2002, volume 31, pages 1289 ^ 1298

DOI:10.1068/p3421

How do the internal details of the object contribute to recognition? Hidemichi Mitsumatsu, Kazuhiko Yokosawa

Department of Psychology, Graduate School of Humanities and Sociology, University of Tokyo, Hongo, Bunkyo-ku, Tokyo 113-0033, Japan; e-mail: [email protected] Received 25 August 2001, in revised form 17 March 2002; published online 16 October 2002

Abstract. Object recognition becomes difficult when the main axis of the object is foreshortened. It has previously been reported that this so-called foreshortened disadvantage is larger when the silhouette of the object is presented than when the line drawing of the object is presented. The pronounced foreshortened disadvantage in silhouette recognition indicates that the internal details of the object, which are absent in the silhouette, provide useful information, particularly when the main axis is foreshortened. But the role of these internal details remains controversial. One account for the pronounced disadvantage is that the internal details contribute to the derivation of the main axis. The other account is that internal details provide the distinctive features that are directly matched to the object represented in memory. The aim in the present study was to determine which of these two explanations best accounts for the differential foreshortened disadvantage between line drawings and silhouettes. To reduce the uncertainty regarding the axis orientation, a 3-D arrow indicating the orientation of the main axis was presented as a cue before the object itself was presented. As a result, the difference in the foreshortened disadvantage between silhouettes and line drawings disappeared. This indicated that the pronounced foreshortened disadvantage for silhouettes was caused by a lack of axis information. In other words, the internal details provided the information necessary for axis derivation when the axis was foreshortened.

1 Introduction Many studies have reported that the presentation angle (view) of an object affects its recognition. Such studies are of two types: studies in which an object is presented once during the experimental session, and those in which an object is presented twice, with the effects of the first presentation on the object recognition in the second presentation being examined. In the initial encounter with an object, recognition becomes difficult relative to the degree of foreshortening of the main axis of the object (Humphrey and JolicÝur 1993; Lawson and Humphreys 1998, 1999). The main axis is the natural axis of symmetry or elongation of the overall 3-D shape of the object. In the second encounter with the object, on the other hand, the difficulty of recognition depends on the difference in presentation angle between the first and second encounters. Recognition becomes more difficult as the difference in the presentation angles becomes larger (eg Lawson and Humphreys 1998, 1999; Tarr 1995; Tarr et al 1998). Thus it is not difficult to recognize a foreshortened object in the second encounter if the object was also foreshortened in the first encounter. Object recognition is view-dependentöie dependent on the presentation angleö irrespective of whether the view is manipulated by 2-D or 3-D rotation. The latter involves the accretion and deletion of the constituent parts of the object from the view, whereas 2-D rotation does not. The accretion and deletion of the parts of the object in 3-D rotation complicates the understanding of the view-dependency of the object recognition. So far, there are two main ways to account for the foreshortened disadvantage: either the foreshortened view obscures the distinctive features needed to identify the object, or the foreshortened view is simply less familiar than other views. The effect of the view on recognition is called the view-dependent effect, or the foreshortened disadvantage in the case of the initial encounter. The view-dependent

1290

H Mitsumatsu, K Yokosawa

effect has been the focus of the debate between two approaches to object recognition. In the part-based approach (Biederman 1987; Marr 1982), the object is recognized as the sum of its constituent parts. The 3-D shape of the parts is defined by taking a base of particular shape (eg a circle) and sweeping it along the main axis of each part. The axes provide information on the orientation, size , and location of the parts, and help to define the object-centered coordinate system. By transforming the viewer-centered coordinate system to the object-centered coordinate system, the objects can be recognized independent of view. When the axis is foreshortened, the derivation of the axis becomes difficult, which makes recognition difficult (Marr 1982). On the other hand, in the view-based approach (Poggio and Edelman 1990; Rock and DiVita 1987; Tarr and Bu«lthoff 1995), the object is represented as a set of 2-D templates, one for each view. Object recognition is achieved by directly matching the input image to the stored templates. The disparities of view between the input and the stored templates incur a cost for compensation, resulting in a view-dependent effect. The existence of a view-dependent effect is often taken as evidence against the part-based approach, because the part-based approach requires the introduction of an object-centered coordinate system to obtain view-invariant recognition. However, the mere presence of a view-dependent effect does not actually constitute compelling evidence against a part-based approach, since the part-based approach admits a view-dependent effect to some extent. For example, Biederman and Bar (2000) suggested that the perceptibility of the parts differs across views, resulting in a view-dependent effect. If the perceptibility of the parts is low from a particular viewpoint, recognition might well be difficult from that viewpoint (for a detailed discussion see Biederman and Bar 1999, 2000; Hayward and Tarr 2000). Moreover, Bar (2001) proposed that the view-dependent effect reflected accessibility to the representation rather than the representation per se. Thus, the mere existence of a view-dependent effect does not in itself validate or invalidate either approach. Is the derivation of the main axis really useful in recognition? There is a previous study of initial object encounters that is relevant to this problem (Lawson and Humphreys 1999). These authors reported that the foreshortened disadvantage for silhouetted objects, in which the internal details were obscured, was larger than that for line-drawn objects. They provided two possible explanations for the differential foreshortened disadvantage between silhouetted and line-drawn objects. In the first explanation, the internal details that are lacking in silhouettes are used in line drawings to assign the main axis of the object in the foreshortened view. When the main axis is less foreshortened, it can be readily derived simply from the outline of the image (Marr 1982), but when it is more foreshortened, internal details are needed to assist in its derivation. The second possible explanation provided by Lawson and Humphreys (1999) is that the internal details do not aid in the derivation of the axis, but rather allow direct recognition by serving as distinctive features (Warrington and James 1986). Distinctive features are clusters of visual contours unique to a particular object. The object recognition can thus be achieved without derivation of the main axis. Instead, object recognition is performed by directly matching the distinctive features with those of the representation of the object stored in memory. In this view-based approach, the foreshortened disadvantage is caused by the less distinctive features in the foreshortened view. Several studies have shown that recognition of silhouettes is virtually as efficient as recognition of shaded images or line drawings when the object is less foreshortened or viewed from a canonical vantage point (Hayward 1998; Lawson and Humphreys 1999; Newell and Findlay 1997). Thus, when the object is less foreshortened, distinctive features in the internal details of line drawings, if any, are not particularly useful, because the outline of the less foreshortened view is sufficiently distinctive. Internal features become useful when the object is foreshortened. In the

The contribution of internal details to object recognition

1291

case of silhouettes, the foreshortened disadvantage becomes pronounced because the outline is no longer distinctive. And, in the case of line drawings, the foreshortened disadvantage is moderate because distinctive features present in the internal details of the foreshortened view can compensate, to some extent, for the lack of distinctive features in the outline. Our aim in the present study was to determine which of the two above explanations best accounts for the differential foreshortened disadvantage between line drawings and silhouettes. This was accomplished by presenting a 3-D arrow that cued the viewer to the orientation of the main axis of the soon-to-be-presented object. If the presence of the arrow reduced the foreshortened disadvantage for silhouettes, then it would follow that internal details are useful for deriving the main axis. On the other hand, if the presence of the arrow did not reduce the foreshortened disadvantage for silhouettes, then it follows that the internal details are not used to derive the main axis, but rather serve as distinctive features for object recognition. 2 Experiment We conducted an experiment to determine which of the two above explanations best accounts for the differential foreshortened disadvantage between line drawings and silhouettes. The 3-D arrow that indicated the orientation of the main axis was presented before the object image. All of the pictures used in the present study depicted objects with either a completely symmetrical 3-D shape or a symmetrical main part at least. In the present study the main axis was defined as the axis of symmetry. A word ^ picture verification task was employed as the object-recognition task. 2.1 Methods 2.1.1 Participants. Forty Japanese adults (thirty-five men and five women; mean age 21 years) participated in the experiment. All participants had normal or corrected-tonormal vision, and all participants were blind to the experimental hypothesis. 2.1.2 Stimuli. A set of two views of 80 familiar objects was produced, with one view in which the objects were foreshortened, and the other view in which they were less foreshortened. The two views of each object were separated by a 308 horizontal rotation in depth. The foreshortened view was oriented so that the object was depicted frontally. In the pilot study, five judges were asked which view was the frontal view for each of the 80 objects. For each of the 80 objects, at least four judges agreed that the foreshortened view presented in the present study was the frontal view. The symmetrical axis of many of the objects coincided with the axis of elongation. In selecting the object and its view, we gave precedence to the axis of symmetry over the axis of elongation, because the foreshortened views of some of the elongated objects, eg a penguin or unicycle, resulted in images that were considered too unusual. Hence, the view of the objects in the present study was defined relative to the symmetrical axis. In the foreshortened view, the orientation of the axis was parallel to the line of sight of the viewer. The less-foreshortened view was created by rotating the object 308 to the left in depth from the foreshortened view (see figure 1). The stimuli were produced by tracing the rendered images of the object models. The images were taken from a slightly elevated angle of 158 or 308 above the horizontal plane on which the objects rested. This angle was kept constant during the depth rotation of the objects. The elevated angle was set so that the effects of foreshortening were not too severe for recognition. The size of each image fit within a 7 deg by 7 deg square. The silhouettes were created by blacking-out the interior of the line drawings. Thus, the outline of the line drawing and its corresponding silhouette were identical; the silhouettes lacked any internal details. The arrows (approximately 4 deg by 4 deg)

1292

H Mitsumatsu, K Yokosawa

Figure 1. Examples of stimuli. The two lefthand columns illustrate line drawings and the two righthand columns illustrate the corresponding silhouettes. Both for the line drawings and for the silhouettes, the leftmost column shows the less-foreshortened view, and the rightmost column shows the foreshortened view.

indicating the orientation of the axes were rendered images shaded in gray scale. In total, four images of arrows were created. Two arrows indicated the orientation of the foreshortened axis: one for the 158 elevation, the other for the 308 elevation. The other two arrows indicated the orientation of the less-foreshortened axis: one for the 158 elevation, the other for the 308 elevation. 2.1.3 Design. The participants were divided into two groups (twelve participants in each group). One group saw only the silhouettes, and the other group saw only the line drawings. Each participant saw each of the 80 objects only once. Thus, there were 80 trials in the experiment. The 80 objects were divided into two sets: one for the 40 match trials, the other for the 40 mismatch trials (see Appendices A and B). For the match trials, the word named the object subsequently presented. For the mismatch trials, in order that the word and presented object would be discriminable, the authors chose words so that the objects they referred to were at least somewhat visually dissimilar, according to the authors' subjective criterion, from the objects subsequently presented. In half of the 80 trials, the foreshortened views of the objects were presented, and the less-foreshortened views were presented in the other half. In half of the 40 trials with foreshortened views, and in half of the 40 trials with less-foreshortened views, the 3-D arrow was presented before the object. The 3-D arrow was not presented in the other half of the 40 trials, respectively. The direction of the arrow indicated the direction that the object was facing. For example, if the arrow was pointing toward the viewer, the frontal view of the object was given and the back view could not be seen. For each participant, whether the object was presented in the foreshortened or lessforeshortened view and whether or not the arrow was presented were determined randomly by custom software.

The contribution of internal details to object recognition

1293

2.1.4 Apparatus and procedure. A Macintosh computer was used to display the stimuli and record the responses. The experiment lasted about 10 minutes. Figure 2 illustrates the sequence of the stimulus presentation. On the initial screen for each trial, a verbal description of whether the arrow would be presented and the name of the object were given. The name was written in Japanese. The size of a single character was approximately 1.5 deg by 1.5 deg. The participant pressed the space key to erase the initial screen. After a 500 ms pause, the 3-D arrow was presented for 1500 ms in the arrow-present trials. For the arrow-absent trials, a plus sign was presented instead of the arrow. The participants were told that the orientation of the arrow indicated the orientation of the symmetrical axis of the object, and were encouraged to use this information to help them complete the verification task. The image of the object was presented immediately after the arrow or plus sign, and remained on the screen until the response. The participants were asked to press the Z key when the name and the presented object matched and to press the M key when they mismatched. They were encouraged to respond as quickly and accurately as possible. After responding, participants were given feedback of whether or not their response was correct. This was done in order to assist them in performing the task with proper criterion, since the shapes of the silhouettes were often ambiguous. Before the experimental session, the participants were shown a list of the 80 words that would be presented in the experiment. This was done in the hope of reducing the load of word recognition and shifting the participants' focus to the object identification. Sixteen practice trials were given. The objects presented in the practice trials were not presented in the experimental session.

until response time

1500 ms arrow present car

arrow present car

press space key after reading the description

Figure 2. The sequence of stimulus presentation. After reading the description on the initial screen, the participants pressed the space key then, after 500 ms, the 3-D arrow appeared for 1500 ms. Then the object was presented and remained on the screen until a response was given. The descriptions on the initial screen were written in Japanese.

2.2 Results Match trials and mismatch trials were analyzed separately. The analyses by participants and by items were reported as F A and F B , respectively. Reaction times (RTs) exceeding 5 s were eliminated as outliers. RT analysis was conducted for correct trials only. 2.2.1 Match trials. A three-way ANOVA (stimulus type6view6arrow presence) revealed a three-way interaction (F1A, 38 ˆ 5:7, p 5 0:05; F1B, 39 ˆ 4:4, p 5 0:05). Then, we conducted

1294

H Mitsumatsu, K Yokosawa

two-way ANOVAs (stimulus type6view) separately for the axis-absent and axis-present conditions. Mean RTs for the axis-absent and axis-present conditions are given in figures 3a and 3b, respectively. For the axis-absent condition, the ANOVA (stimulus type6view) revealed a main effect of view (F1A, 38 ˆ 14:3, p 5 0:01; F1B, 39 ˆ 4:3, p 5 0:05). There was also an interaction (F1A, 38 ˆ 4:2, p 5 0:05; F1B, 39 ˆ 4:7, p 5 0:05). For the axis-present condition, the two-way ANOVA (stimulus type6view) revealed a main effect of view (F1A, 38 ˆ 10:8, p 5 0:05; F1B, 39 ˆ 6:5, p 5 0:05). However, there was no interaction (F1A, 38 ˆ 0:005, p 4 0:9; F1B, 39 ˆ 0:05, p 4 0:8). Two-way ANOVAs (arrow presence6view) were conducted separately for the line-drawing and silhouette conditions. For line drawings, the interaction was not significant (F1A, 19 ˆ 0:2, p 4 0:6; F1B, 39 ˆ 0:3, p 4 0:5). For silhouettes, the interaction was significant in the by-items analysis (F1A, 19 ˆ 3:2, p 5 0:09; F1B, 39 ˆ 4:1, p 5 0:05). 1200

less-foreshortened foreshortened

Reaction times=ms

1100 1000 900 800 700 600 500

(a)

Line drawing

Silhouette

Line drawing

(b)

Silhouette

Figure 3. The results of the match trials: (a) the arrow-absent condition, and (b) the arrow-present condition. Error bars are for standard errors.

For the analysis of error rates, a three-way ANOVA showed no three-way interaction (F1A, 38 ˆ 0:1, p 4 0:7; F1B, 39 ˆ 2:4, p 4 0:1). There was an interaction of stimulus type 6view in the by-participants analysis (F1A, 38 ˆ 4:3, p 5 0:05; F1B, 39 ˆ 0:1, p 4 0:8). That is, the foreshortened disadvantage for silhouettes (8.5% versus 22%) was larger than that for line drawings (3.8% versus 11.2%). 2.2.2 Mismatch trials. A three-way ANOVA (stimulus type6view6arrow presence) revealed a three-way interaction (F1A, 38 ˆ 7:2, p 5 0:05; F1B, 39 ˆ 4:2, p 5 0:05). We then conducted two-way ANOVAs separately for the axis-absent and axis-present conditions. Mean RTs for the axis-absent and axis-present conditions are given in figures 4a and 4b, respectively. For the axis-absent condition, the ANOVA (stimulus type6view) revealed a main effect of view (F1A, 38 ˆ 12:1, p 5 0:01; F1B, 39 ˆ 17:0, p 5 0:001). An interaction was also revealed (F1A, 38 ˆ 5:7, p 5 0:05; F1B, 39 ˆ 4:4, p 5 0:05). For the axis-present condition, a two-way ANOVA (stimulus type6view) revealed a main effect of view (F1A, 38 ˆ 8:0, p 5 0:01; F1B, 39 ˆ 9:9, p 5 0:01). However, the interaction was not significant (F1A, 38 ˆ 0:62, p 4 0:4; F1B, 39 ˆ 0:007, p 4 0:9). Two-way ANOVAs (arrow presence6view) were conducted separately for the line-drawing and silhouette conditions. For line drawings, the interaction was not significant (F1A, 19 ˆ 1:7, p 4 0:2; F1B,;39 ˆ 0:1, p 4 0:7). For silhouettes, the interaction was significant (F1A, 19 ˆ 5:1, p 5 0:05; F1B, 39 ˆ 4:9, p 5 0:05). For the analysis of error rates, a three-way ANOVA showed that none of the interactions was significant. The main effect of view was significant (F1A, 38 ˆ 0:62, p 4 0:4; F1B, 39 ˆ 0:007, p 4 0:9).

The contribution of internal details to object recognition

1295

1200

less-foreshortened foreshortened

Reaction times=ms

1100 1000 900 800 700 600 500

Line drawing

Silhouette

Line drawing

Silhouette

(a) (b) Figure 4. The results of the mismatch trials: (a) the arrow-absent condition, and (b) the arrowpresent condition. Errors bars are for standard errors.

3 Discussion In the present study, the RT patterns for the match trials were similar to those for the mismatch trials. The arrow-absent condition in the present experiment replicated the finding of Lawson and Humphreys (1999) that the foreshortened disadvantage for silhouettes was larger than that for line drawings. In contrast with the arrow-absent condition, the arrow-present condition showed that the foreshortened disadvantage for silhouettes was no different from that for line drawings. These results revealed that the presence of the arrow reduced the foreshortened disadvantage for silhouettes. Thus, it seems that the severe foreshortened disadvantage for silhouettes is caused by the difficulty of deriving the axis in the foreshortened view, and is not because distinctive features are obscured. The analysis of error rates did not show an effect of the arrow on recognition. Thus, the effect of the arrow could not be discussed with respect to error rates. We interpret the effect of the arrow as evidence that the internal details of the object that are lacking in silhouettes are useful for derivation of the axis. However, it might also be argued that the effect of the arrow is unrelated to the axis derivation. If the less-foreshortened view was the preferred view for a given participant, then that participant might have imagined the less-foreshortened view of the object upon reading its name (Palmer et al 1981). If the viewers properly imagined the foreshortened view when the foreshortened arrow was presented, by mentally rotating the imagined lessforeshortened object, for example, the effect of the arrow would have nothing to do with the derivation of the axis. That is, matching the input view to the imagined view is the process that view-based approaches have proposed. If the imagined view is used for recognition, the object is represented in the viewer-centered coordinate system. However, the diminishing of the foreshortened disadvantage in the mismatch trials ran counter to this interpretation. The shape of the object that corresponded to the word in the mismatch trials would not match the actually depicted object, no matter how closely the participants tried to match the imagined view to the actually presented view. Thus, the arrow should be used to derive the axis rather than to imagine the view of a not-yet-presented object. Moreover, several studies have reported that humans have extreme difficulty in imagining the rotated view from the input image (Pani 1993; Pani and Dupree 1994; Parsons 1995). These studies lend support to our interpretation of the results. Another issue that should be addressed is that the effect of the arrow may be to direct spatial attention to the appropriate axis. We conjecture that this is not the case.

1296

H Mitsumatsu, K Yokosawa

In our experimental design, the arrow-absent condition and arrow-present condition trials were mixed in a block. Thus, the participants were aware of the strategy of focusing on the orientation of the main axis even in the arrow-absent condition. Nonetheless, the results showed a pronounced foreshortened disadvantage for silhouettes in the arrow-absent condition. Therefore, the reduction in the foreshortened disadvantage for silhouettes in the arrow-present condition seemed to be caused by the knowledge of the orientation of the main axis, but not by attention directed to the main axis. The results of the present study might run counter to the framework of global-tolocal processing. That is, the arrow as local information preceded the object as global information. Nonetheless, the foreshortened disadvantage was reduced by the presence of the arrow. We infer that the recognition process of the foreshortened view might be the exception to this framework. If the derivation of the main axis is crucial for the object recognition, it might be that global information is preceded by local information such as the internal details in order to obtain information on the orientation of the main axis when the global shape of the foreshortened view lacks the information on the orientation of the main axis. After information on the main axis has been obtained from the internal detail, the focus of the process might be redirected to the global information. If the axis is derived from the internal details of the object, how does this process take place? Though the present study was not designed to answer this question, we can suggest several possibilities. First, symmetrically configured internal details might become a cue for deriving the main axis. For example, if the object has pairs of identical features (eg the headlights of a car), the configurations of the pairs remain symmetrical in the foreshortened view, and the viewer can derive the symmetrical axis. Second, the orientation of lines in the internal details might become a cue for deriving the main axis. For example, if the object is composed of two symmetrical main parts (eg an open book or butterfly), the orientation of the junction of the parts coincides with that of the axis of symmetry. Third, the texture gradient of features due to the perspective projection might become a cue for deriving the main axis. In general, visible features become finer when they are placed at a higher position in the image because the viewing angle of the objects is elevated (158 or 308), as it was in the present study. In the foreshortened view, the orientation of the texture gradient was vertical in the projected image, whereas the orientation was oblique (visible features become finer as they are placed from lower right to upper left in the projected image) in the less-foreshortened view. It should be noted that the symmetry of the outline shape was not used to derive the orientation of the main axis. Because the objects presented in the present study had a virtually symmetrical 3-D shape, the outline shape was symmetrical in the foreshortened view, but not symmetrical in the less-foreshortened view. The viewers could have used this information to derive the orientation of the main axis, with the result that silhouettes would not have had a pronounced foreshortened disadvantage. In fact, however, a pronounced foreshortened disadvantage was observed, indicating that the symmetry of the outline shape was not useful in deriving the orientation of the main axis. 4 Conclusion The present study was conducted to determine which of two commonly given theoriesöie the main-axis and the distinctive-feature accountsöbest explains the differential foreshortened disadvantage between line drawings and silhouettes. We preceded the presentation of the object with an arrow cueing the orientation of the main axis. As a result, the difference in the foreshortened disadvantage between silhouettes and line drawings disappeared. The results indicated that the severe foreshortened disadvantage

The contribution of internal details to object recognition

1297

for silhouettes was caused by a lack of information as to the orientation of the main axis. In other words, the internal details in the line drawings provided the information necessary for the derivation of the main axis when the axis was foreshortened. Acknowledgments. This research was supported by a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (No. 12871012 and No. 13224021; awarded to Kazuhiko Yokosawa). References Bar M, 2001 ``Viewpoint dependency in visual object recognition does not necessarily imply viewer-centered representation'' Journal of Cognitive Neuroscience 13 793 ^ 799 Biederman I, 1987 ``Recognition-by-components: A theory of human image understanding'' Psychological Review 94 115 ^ 147 Biederman I, Bar M, 1999 ``One-shot viewpoint invariance in matching novel objects'' Vision Research 39 2885 ^ 2899 Biederman I, Bar M, 2000 ``Differing views on views: response to Hayward and Tarr (2000)'' Vision Research 40 3901 ^ 3905 Hayward W G, 1998 ``Effects of outline shape in object recognition'' Journal of Experimental Psychology: Human Perception and Performance 24 427 ^ 440 Hayward W G, Tarr M J, 2000 ``Differing views on views: comments on Biederman and Bar (1999)'' Vision Research 40 3895 ^ 3899 Humphrey G K, JolicÝur P, 1993 ``Visual object identification: Some effects of image foreshortening, monocular depth cues, and visual field on object identification'' Quarterly Journal of Experimental Psychology 46A 137 ^ 159 Lawson R, Humphreys G W, 1998 ``View-specific effects of depth rotation and foreshortening on the initial recognition and priming of object'' Perception & Psychophysics 60 1052 ^ 1066 Lawson R, Humphreys G W, 1999 ``The effects of view in depth on the identification of line drawings and silhouettes of familiar objects: normality and pathology'' Visual Cognition 6 165 ^ 195 Marr D, 1982 Vision (San Francisco, CA: W H Freeman) Newell F N, Findlay J M, 1997 ``The effect of depth rotation on object identification'' Perception 26 1231 ^ 1257 Palmer S, Rosch E, Chase P, 1981 ``Canonical perspectives and the perception of objects'', in Attention and Performance IX Eds J Long, A Baddeley (Hillsdale, NJ: Lawrence Erlbaum Associates) pp 135 ^ 151 Pani J R, 1993 ``Limits on the comprehension of rotational motion: mental imagery of rotation with oblique components'' Perception 22 785 ^ 808 Pani J R, Dupree D, 1994 ``Spatial reference systems in the comprehension of rotational motion'' Perception 23 929 ^ 946 Parsons L M, 1995 ``Inability to reason about an object's orientation using an axis and angle of rotation'' Journal of Experimental Psychology: Human Perception and Performance 21 1259 ^ 1277 Poggio T, Edelman S, 1990 ``A network that learns to recognize three-dimensional objects'' Nature 343 263 ^ 266 Rock I, DiVita J, 1987 ``A case of view-centered perception'' Cognitive Psychology 19 280 ^ 293 Tarr M J, 1995 ``Rotating objects to recognize them: A case study on the role of viewpoint dependency in the recognition of three-dimensional objects'' Psychological Bulletin & Review 2 55 ^ 82 Tarr M J, Bu«lthoff H H, 1995 ``Is human object recognition better described by geon-structuraldescriptions or multiple views? Comment on Biederman and Gerhardstein (1993)'' Journal of Experimental Psychology: Human Perception and Performance 21 1494 ^ 1505 Tarr M J, Williams P, Hayward W G, Gauthier I, 1998 ``Three-dimensional object recognition is viewpoint dependent'' Nature Neuroscience 1 275 ^ 277 Warrington E K, James M, 1986 ``Visual object recognition in patients with right hemisphere lesions: axes or features?'' Perception 15 355 ^ 366

1298

H Mitsumatsu, K Yokosawa

APPENDIX A Table A1. The 40 objects that matched the words presented. Airplane Bed Blow-dryer Book Bread Butterfly Camel Car Clarinet Desk

Dog Dustpan Eagle Elephant Frog Helicopter House Motorbike Pan Penguin

Phone Piano Rabbit Sailboat Scooter Ship Shoe Sofa Stapler Table

Tank Teapot Tiger Truck Trumpet Tube Turtle TV set Violin Watering can

APPENDIX B Table B1. The 40 mismatched words and the objects presented. Object depicted

Word presented

Object depicted

Word presented

Armadillo Banana Bee Calculator Chest Computer Crocodile Desk lamp Flamingo Giraffe Hammer Helmet Horse Kangaroo Key Lipstick Lobster Megaphone Office chair Ostrich

Cat Radish Beetle Newspaper Pot Oven Glow fly Microscope Chicken Wolf Axe Crown Hippopotamus Duck Knife Oboe Crab Bat Wagon Swan

Padlock Paperclip Racquet Rhino Saxophone Scissors Sewing machine Shark Snail Space shuttle Spectacles Syringe Telescope Toaster Tractor Unicycle Video camera Watch Wheelchair Xylophone

Bucket Tape Spoon Deer Broom Pliers Suitcase Seal Socks Whale Binoculars Pencil Peacock Rugby ball Dump truck Spade Iron Tape measure Wheelbarrow Organ

ß 2002 a Pion publication