Vision Research 44 (2004) 2779–2791 www.elsevier.com/locate/visres A new cue to ﬁgure–ground coding: top–bottom polarity Johan Hulleman *, Glyn W. Humphreys School of Psychology, University of Birmingham, Birmingham B15 2TT, UK Received 8 March 2004; received in revised form 7 June 2004 Abstract We present evidence for a new ﬁgure–ground cue: top–bottom polarity. In an explicit reporting task, participants were more likely to interpret stimuli with a wide base and a narrow top as a ﬁgure. A similar advantage for wide-based stimuli also occurred in a visual short-term memory task, where the stimuli had ambiguous ﬁgure–ground relations. Further support comes from a ﬁgural search task. Figural search is a discrimination task in which participants are set to search for a symmetric target in a display with ambiguous ﬁgure–ground organization. We show that ﬁgural search was easier when stimuli with a top–bottom polarity were placed in an orientation where they had a wide base and a narrow top, relative to when this orientation was inverted. This polarity eﬀect was present when participants were set to use color to parse ﬁgure from ground, and it was magniﬁed when the participants did not have any foreknowledge of the color of the symmetric target. Taken together the results suggest that top–bottom polarity inﬂuences ﬁgure–ground assignment, with wide base stimuli being preferred as a ﬁgure. In addition, the ﬁgural search task can serve as a useful procedure to examine ﬁgure–ground assignment. 2004 Elsevier Ltd. All rights reserved. Keywords: Perceptual organization; Figure–ground segregation 1. Introduction In 1917, Rubin published his observations about the alternative perceptual organizations of his famous ﬁgure-vase stimulus. On the one hand, the stimulus can be interpreted as two faces looking at each other in front of a rectangle; on the other it can also be seen as a vase in front of the same rectangle, with the color of the rectangle now changed. In the many years that have since passed, research has indicated that ﬁgure–ground organization is inﬂuenced by a variety of cues including: Symmetry, convexity, surroundedness, area, familiarity and spatial frequency (see Palmer, 1999 for a discussion). Recently, Vecera, Vogel, and Woodman (2002) have added Ôlower regionÕ to this list. They demonstrated a * Corresponding author. Tel.: +44 121 4143679. E-mail address: [email protected] (J. Hulleman). 0042-6989/$ - see front matter 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.visres.2004.06.012 bias in ﬁgure–ground organization such that the bottom half of a whole visual display tended to be taken as the ﬁgure, and the top half as the ground. One of the problems in studying ﬁgure–ground organization is in choosing a task that reﬂects ﬁgure– ground coding, but that is not aﬀected by other factors. In the vast majority of studies, observers have simply been asked to record what they saw as a ﬁgure (an explicit reporting task). However, this introspective process can be aﬀected by many factors. For instance, with ambiguous stimuli there can be individual biases in maintaining or in wishing to switch from one organization to another, which can have quite dramatic eﬀects on performance (e.g. Strüber & Stadler, 1999). An alternative procedure, less dependent on introspective biases, has been to use visual short-term memory tasks (VSTM-task, introduced by Driver & Baylis, 1996). For example, observers may be presented with an ambiguous ﬁgure–ground stimulus for a certain period 2780 J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 (the study display), followed by a test display containing one of the regions from the ambiguous study display. The idea is that the ﬁgural region will have a stronger memory trace, so that the reaction times (RTs) will be faster when the subsequent region matches the perceived ﬁgure rather than the perceived ground in the study display (see Vecera et al., 2002, for an example). This method has become something of a gold standard in ﬁgure–ground research. Even this method, though, might not be without its problems. For instance, it is possible that eﬀects in short-term memory tasks are caused by properties of the second, test display, rather than the original study display containing the ambiguous ﬁgure–ground stimulus. So, it seems that, in order to establish that a certain cue inﬂuences ﬁgure–ground assignment, the prudent approach would be to use all the methods that are available, rather than to rely on a single approach. For instance, Vecera et al. (2002) used both explicit reporting tasks and the VSTM-method to establish that lower region is a ﬁgure–ground cue. In the present study we use both of the above procedures, plus also a new method (‘‘ﬁgural search’’) to establish another new cue for ﬁgure–ground assignment: top–bottom polarity. We use the evidence to argue both for the role of this new cue, and for the utility of ﬁgural search as a means to examine ﬁgure–ground coding. We employed abstract shapes, depicted within a band pattern stimulus as used by Metzger (1936). Examples are shown in Fig. 1. The abstract shapes were constructed so that they had two long horizontal elements either at the base, and two short horizontal elements at the top (wide base) or two short horizontal elements at the base and the two long horizontal elements at the top (wide top). This resulted in a strong top–bottom polarity (see the Method). 2. Experiment 1: direct report In Experiment 1 we used an explicit report task to provide direct evidence on any eﬀect of top–bottom polarity on ﬁgure–ground coding. We presented a group of participants with ambiguous ﬁgure–ground displays. There were four displays (see Fig. 1), with the wide base regions being black for two displays and white for two displays (and vice versa for the wide top regions). For each display, participants had to decide whether the white or the black regions were seen as the ﬁgure. 2.1. Method The study was run in a classroom testing situation. There were 55 participants (38 female, 17 male), all undergraduates in Psychology at the University of Birmingham (aged between 19 and 22). The four displays Fig. 1. The four stimuli used in Experiment 1. were shown one at a time, projected onto a large screen at the front of the classroom. The order of the displays was randomly determined. All the regions of the band patterns were asymmetric. Participants were not allowed to confer, and they made a written response to each display. In an initial pilot experiment, 22 observers were asked to view a pair of stimuli from the set used here, but with just a single element from the band patterns used to represent the Ôwide topÕ and Ôwide baseÕ stimulus in each pair. A forced choice decision was then required as to whether the Ôwide topÕ or the Ôwide baseÕ stimulus was upright. A set of 8 pairs was shown. Twenty participants chose all 8 stimuli with a Ôwide baseÕ as upright. One participant chose 1 stimulus as upright from the Ôwide topÕ items. One other participant chose 3 stimuli with a wide top as upright. Thus overwhelmingly observers chose the Ôwide baseÕ items as upright. 2.2. Results The percentages of the participants who reported the wide base shapes as ﬁgure were 89%, 76%, 84% and 71% J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 respectively, for the four displays. Tested binomially, all four percentages diﬀer signiﬁcantly from 50% (all pÕs < 0.001). 2.3. Discussion There was a substantial and reliable bias in the explicit report of which part of an ambiguous display was perceived as ÔﬁgureÕ. Participants were biased to interpret the wide base regions as ﬁgural. This occurred irrespective of whether these regions were white or black. Note that all the regions were asymmetric, and that the participants were not engaged in a symmetry detection task. Therefore, the results cannot be attributed to eﬀects of top–bottom polarity on symmetry perception, nor on symmetry biasing the ﬁgure–ground assignment. As pointed out in the Introduction, explicit reporting methods need to be supplemented by more implicit methods, because the deliberate nature of explicit reporting tasks leaves open the possibility of extraneous factors. In the remainder of the experiments, we therefore used implicited methods, to provide converging evidence that top–bottom polarity acts as a ﬁgure–ground cue. In Experiment 2, we used the VSTM-task and in Experiments 3 and 4 we used a new implicit task, the ﬁgural search task. 3. Experiment 2: matching shapes in VSTM In the VSTM task, participants are presented with a display with an ambiguous ﬁgure–ground relationship between black and white regions. This is followed by a pair of test shapes (unambiguous ﬁgures against the overall ground; see Fig. 2). The task is to decide which Fig. 2. Sequence of displays in Experiment 2. After the presentation of a ﬁxation dot, the study stimulus was presented for 200 ms. After a 500 ms blank interval, the test display appeared. The test displays remained visible until response, or until 15 s had elapsed. 2781 of the test shapes had a bounding contour that matched the border between the black and white regions. Performance is better if the contour of the matching test shape corresponds to the boundary of the ﬁgural region in the initial display (Driver & Baylis, 1996). Here we would expect a beneﬁt when the contour matches the boundary of the wide base stimulus, rather than the wide top stimulus. 3.1. Method 3.1.1. Participants Twenty-two participants (20 female, 2 male) were recruited either in return for course credit or for a small payment. All had corrected or corrected-to-normal vision, and they were unaware of the purpose of this experiment. 3.1.2. Apparatus A Pentium III PC controlled the experiment and presented the stimuli on a 17 in. VGA-monitor, in 800 · 600 graphics mode. 3.1.3. Stimuli The participants looked unrestrained at the stimuli from standard viewing distance (around 50 cm). The stimulus used in the study display contained two juxtaposed shapes, one black and one white (3.7 · 3.7), presented on a gray background. The test display contained two dark gray shapes (2.1 · 3.7). Examples are shown in Fig. 6. The study stimulus appeared at the center of the screen. The vertical size of an individual shape was 3.7. A half of a shape consisted of 10 arms, each arm having a thickness of 0.37. The maximum width of an arm was 1.1. The minimum width was 0.2. The shapes were always asymmetric. The shapes were rotationally symmetric. As a result, a shape and its 180 rotated version ﬁt each other like a jig-saw when juxtaposed. This yields a natural control for non-speciﬁc eﬀects on the ﬁgure– ground assignment, because the shape preferred as a ﬁgure and the shape preferred as a ground are identical. In this experiment we used 15 diﬀerent shapes. Each shape was used 16 times: The wide-base shape could be on the left or on the right, the wide-base shape could be black or could be white, and the test display could contain a ﬁgural match for the wide-base shape or the wide-top shape, with the correct answer either on the left or the right. 3.1.4. Procedure The participants were tested individually in a session that lasted approximately 20 min. After receiving the instruction for the task, participants performed a practice block of 16 trials. When they felt at ease with the task, they would start the experimental blocks, otherwise they would get another practice block of 16 trials. 2782 J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 A trial consisted of a ﬁxation circle presented in the center of the screen for 500 ms. After the ﬁxation circle disappeared, the stimulus was presented for 200 ms. After a blank display that lasted 500 ms, the test display appeared, and remained on the screen until the participant responded or 30,000 ms had passed. After the response, there was an inter-trial interval of 1000 ms, before a new trial would start (see Fig. 2). The task of the participants was to decide which of the two shapes presented in the test display had a bounding contour that corresponded to the border between the black and the white shape seen in the study display. If it was the shape on the left, they had to press the ÔnÕ on the keyboard as quickly as possible. For the shape on the right the Ô/Õ key was used. 3.1.5. Design There were two within-subject factors: study position (wide base item left, wide base item right), and ﬁgural match (left item, right item). The factors were fully crossed (see Fig. 3 for examples of the four possible com- binations). If top–bottom polarity biases ﬁgure–ground assignment, we would expect an interaction between these two factors: if during the study display the wide base item appears on the left, the ﬁgural match of the left item should be faster, but the ﬁgural match of the right item should be faster if the wide base item appears on the right during the study display. In total, there were 240 trials. The experiment was subdivided into blocks of 64. Between every block there was a self-paced break for the participant. If a participant made an error, the error trial was retaken somewhere in the sequence of trials. Each error resulted in an extra trial. 3.2. Results The results are shown in Fig. 4. The data of one participant had to be discarded, because of an excessive error rate (proportion correct: 0.46). The data of the remaining 21 participants were entered into the analysis. Responses greater or less than 2.5 SDs from the cell mean (study position · ﬁgural match) for each partici- Fig. 3. Examples of the stimuli used in Experiment 2. (A) Stimulus with the wide base stimulus on the left and the wide top stimulus on the right: (i) test display with a ﬁgural match for the right item and (ii) test display with a ﬁgural match for the left item. (B) Stimulus with the wide base stimulus on the right and the wide top stimulus on the left: (iii) test display with a ﬁgural match for the right item and (iv) test display with a ﬁgural match for the left item. J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 Reaction time [msec] 1000 Wide Base Item Left Wide Base Item Right 950 900 850 800 Left Item Right Item Figural Match 0.2 Proportion Error Wide Base Item Left Wide Base Item Right 0.15 0.1 0.05 0 Left Item Right Item Figural Match Fig. 4. Results of Experiment 2. Top panel: reaction times. Bottom panel: error proportions. White: wide base item on the left in the study display. Black: wide base item on the right in the study display. The error bars indicate standard errors. pant were rejected. This resulted in rejection of 2.8% of the data. The error analysis was performed on all of the remaining trials, whereas the reaction time analysis only used the correct trials. Critically, a two-way repeated measures ANOVA (study position · ﬁgural match) on the RT data yielded a signiﬁcant interaction between study position and ﬁgural match, F(1,20) = 5.042, p < 0.037. Neither of the main eﬀects was signiﬁcant (both pÕs > 0.49). A similar ANOVA on the error-data did not yield any signiﬁcant eﬀects. All p-values were 0.66 or above. 3.3. Discussion The results of Experiment 2 support those obtained in Experiment 1. The interaction between study position and ﬁgural match indicates that there was an RT advantage for ﬁgural matches of the wide base region relative to matches contingent on the wide top region. This advantage is consistent with top–bottom polarity aﬀecting ﬁgural assignment in the ambiguous displays. Note that this result cannot be explained by eﬀects of top–bottom polarity on symmetry detection, because all 2783 the shapes here were asymmetric. Also, all RTs are well within the range of times that have previously been accepted to be associated with ﬁgure–ground assignment (e.g. Driver & Baylis, 1996, Vecera et al., 2002). In Experiments 3 and 4 we used a new implicit method: the ﬁgural search task. It combines elements of both the implicit (VSTM) and explicit report methods used in prior studies. The task provides an on-line measure (rather than a memory-based measure) of ﬁgure–ground assignment, 1 but one where (as with the memory tasks) observers are not explicitly required to report what they see as a ﬁgure. We presented observers with displays with ambiguous ﬁgure–ground relations, and asked them to ﬁnd a particular target shape (in our case, a vertically symmetric item). We will call this task ﬁgural search, because it combines elements from ﬁgure– ground research with visual search. If a given cue leads to either an initial or dominant ﬁgure–ground organization, then the target shape revealed by this cue should be easier to ﬁnd than a target shape not revealed by this cue. This then provides an indication that a particular cue inﬂuences ﬁgure–ground assignment. Our expectation, that it will take longer to detect a symmetrical shape when it belongs to the ground, is based on results reported by Baylis and Driver (2001). They showed that is more diﬃcult to detect symmetry when the symmetrical edges belong to two diﬀerent shapes (as is the case when the target shape is seen as belonging to the ground), than when the symmetrical edges belong to a single shape (as is the case when the target shape is seen as belonging to the ﬁgure). 4. Experiment 3: ﬁgural search for a known target color In Experiment 3, participants were set to search for a symmetrical target amongst asymmetrical distractors. The target was also coded by color (either black or white, against a gray background). Under these circumstances, participants should parse ﬁgural from ground regions using color. The stimuli (abstract shapes) could either have a wide base, or a wide top. We ask whether the top–bottom polarity of the stimuli aﬀected performance, even when ﬁgure–ground organization should be based on color. We note that, although the task was symmetry detection, there was only one symmetrical shape in a display. Hence, ﬁgure–ground should not be overwhelmingly determined by symmetry-eﬀects of color and top–bottom polarity may thus still emerge (as indeed we observed). 1 With on-line, we mean that the participants make their response while the ambiguous ﬁgure–ground stimulus is still available for inspection. 2784 J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 4.1. Method 4.1.1. Participants Sixteen participants (13 female, 3 male) were recruited either in return for course credit or for a small payment. All had corrected or corrected-to-normal vision, and they were unaware of the purpose of this experiment. 4.1.2. Apparatus A Pentium III PC controlled the experiment and presented the stimuli on a 17 in. VGA-monitor, in 800 · 600 graphics mode. 4.2. Stimuli The participants looked unrestrained at the stimuli from a standard viewing distance (around 50 cm). The stimuli were black and white band patterns (3.7 · 12.4), presented on a gray background. Examples are shown in Fig. 5. The band patterns appeared at the center of the screen. Each band pattern contained eight complete objects and two half objects. Half of them were black, the other half were white. The vertical size of an individual shape was 3.7. A half of a shape consisted of 16 arms, each arm having a thickness of 0.2. The maximum width of an arm was 1.1. The minimum width was 0.2. The arms of a half were paired: the arm at the top end with the arm at the bottom end, the second arm from the top with the second arm from the bottom, etc. The lengths of a pair of arms had to sum up to 1.4. For the wide base stimuli, the halves were constructed in such a way that the two arms at the bottom end of the shape always (nearly) had the maximum width. Consequently, the two arms at the top end always (nearly) had the minimum width. For the wide top stimuli, this was inverted. The lengths of the other arms were randomly chosen to be between the minimum width and the maximum width. Complete objects were created by combining two halves. In the case of an asymmetric object, the halves were constructed independently, for symmetric objects the same half was used twice. An individual band pattern was used four times: Once with the wide base items colored black and the wide top items colored white, once with this color assignment inverted, and both these versions were also presented upside down. 4.2.1. Procedure The participants were tested individually in a session that lasted approximately 40 min. After receiving the instruction for the task, participants performed a practice block of 16 trials. When they felt at ease with the task, they would start the experimental blocks, otherwise they were given another practice block of 16 trials. A trial consisted of a ﬁxation cross presented in the center of the screen for 1000 ms. After the ﬁxation cross disappeared, the band pattern was presented until the participant responded or 15,000 ms had passed. After the response, there was an inter-trial interval of 1000 ms, before a new trial would start. The task of the participants was to look for a vertically symmetric object of a speciﬁed color on the screen. If it was present, they had to push the ÔZÕ or ÔMÕ key on the keyboard as quickly as possible. If it was absent, they had to push the other key (ÔMÕ or ÔZÕ). The keys assigned to present and absent responses were counterbalanced over participants. 4.2.2. Design There were three within-subject factors: top–bottom polarity (wide base, wide top), target presence (symmetric target present, symmetric target absent), and the color of the stimuli (black, white). All the factors were fully crossed. In total, there were 640 trials. The target colors were blocked, resulting in two groups of 320 trials. The groups contained blocks of 80 trials. Between every block there was a self paced break for the participant. If a participant made an error, the error trial was retaken somewhere in the sequence of trials. Each error resulted in an extra trial. After the ﬁrst target color was tested, there was a new training sequence of sixteen trials, to give the participants the opportunity to acquaint themselves with the new target color. 4.3. Results Fig. 5. Examples of displays with ambiguous ﬁgure–ground relations used in Experiments 3 and 4. Top half: target color black; and Bottom half: target color white. The results are shown in Fig. 6. Responses greater or less than 2.0 SDs from the cell mean (top–bottom polarity · target presence · color) for each participant were rejected. This resulted in rejection of 4.2% of the data. The error analysis was performed on all of the remain- J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 4.4. Discussion Reaction Time [msec] 2500 Wide Base 2250 Wide Top 2000 1750 1500 1250 1000 Absent (A) Present Symmetric Target 0.125 Wide Base Proportion Error 0.100 Wide Top 0.075 0.050 0.025 0.000 Absent (B) 2785 Present Symmetric Target Fig. 6. Results of Experiment 3: (A) reaction times and (B) error rates. Black: wide base stimuli. White: wide top stimuli. Left: symmetric target absent. Right: symmetric target present. The error bars indicate standard errors. ing trials, whereas the reaction time analysis only used the correct trials. A three way repeated measures ANOVA (top–bottom polarity · target presence · color) on the reaction time data yielded signiﬁcant main eﬀects for target presence, F(1,15) = 55.9, p < 0.001 and top–bottom polarity, F(1,15) = 15.8, p < 0.002. Present responses were faster than absent, and responses to targets with a wide base were faster than responses to targets with a wide top. There were no other signiﬁcant main eﬀects nor were there any signiﬁcant interactions. A similar ANOVA on the error-data resulted in signiﬁcant main eﬀects for target presence, F(1,15) = 17.9, p < 0.002, and top–bottom polarity, F(1,15) = 17.8, p < 0.002. Moreover, there was a signiﬁcant interaction between target presence and top–bottom polarity, F(1,15) = 8.6, p < 0.011. There were more errors on present trials than on absent trials, and fewer for targets with a wide base than for targets with a wide top. This advantage for wide-base targets was most pronounced on target present trials. For all the other main eﬀects and interactions, the p-values were larger than 0.2. We consider the ﬁgural search task as a form of visual search in which participants have to select a symmetrical target from asymmetric distractors. Olivers and Van der Helm (1998) have shown that visual search for a symmetrical target is ineﬃcient, and the long reaction times and high error rates for the present ﬁgural search task are consistent with this. More interesting is that ﬁgural search was aﬀected by the top–bottom polarity of the target shape, even though the color of the target was pre-speciﬁed so that participants should have been set to organize ﬁgure–ground relations on the basis of color. For the reaction time measure, the eﬀect was quite substantial and equally large on present and absent trials (an overall eﬀect of around 100 ms). In addition to this, there was a high error rate, with targets with a wide top being particularly hard to detect (a miss rate of nearly 10%). However, even though there are more errors on present trials than on absent trials, there are always more errors in the wide top condition than in the wide base condition. So the polarity eﬀect we observe is not caused by speed-accuracy trade-oﬀ. The long reaction times and high miss rate for widetop targets may come about because top–bottom polarity is a powerful factor inﬂuencing ﬁgure–ground organization. To detect a target with a wide top, participants must create and maintain a representation in which the wide top stimuli are coded as ÔﬁgureÕ, and the wide base stimuli as ÔgroundÕ. The data suggest that it is diﬃcult to do this. Symmetry has repeatedly been reported as a determiner of ﬁgure and ground (e.g. Driver & Baylis, 1996; Palmer, 1999). However, the 100 ms diﬀerence between the wide base and the wide top conditions cannot be attributed to the presence or absence of a symmetric target. Both the wide base and the wide top conditions contained the same number of symmetrical targets. If symmetry biased the assignment of ﬁgure and ground, we would have expected to ﬁnd no diﬀerence between the wide base and the wide top conditions. However, other accounts are still possible. For example, it may simply be that visual search through items with a wide top is more diﬃcult than search through items with a wide base, and this is unrelated to ﬁgure– ground segmentation. Several investigators have demonstrated that visual search is inﬂuenced by the perceived orientations of stimuli, with asymmetries favoring oblique over vertical stimuli (Treisman & Souther, 1985) and stimuli in unusual over ÔstandardÕ orientations (Wolfe, 2001). Given the similarities between visual search and our ﬁgural search tasks, it may be that, in Experiment 3, we witness an asymmetry, favoring items with a wide base over items with a wide top (Note that this would actually constitute a case of a standard orientation being favored). The results of Experiments 1 and 2786 J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 2, where we found an eﬀect of top–bottom polarity on ﬁgure–ground assignment with tasks that are unrelated to visual search, suggest it is unlikely that a search assymmetry is the only explanation, but it might be a contributing factor. To test this, in Experiment 4, we included a new manipulation. In one condition we used stimuli with ambiguous ﬁgure–ground relations (as in Experiment 3), but this time we allowed the targets to be either black or white on a trial. Thus color was no longer a reliable cue to ﬁgure–ground organization. If part of the beneﬁt to wide base stimuli in Experiment 3 was because they bias ﬁgure–ground assignment, then even stronger eﬀects of top–bottom polarity could occur here, since the moderating eﬀect of another factor (the color cue) was eliminated. In the second condition, participants were presented with displays in which individual search elements did not have ambiguous ﬁgure–ground relations (for the most part being completely surrounded by the overall gray background). This condition was included because pilot work indicated that the ﬁgural search task without color instruction was quite diﬃcult. We therefore felt it was necessary to include an ÔeasyÕ condition, in order to keep the participants motivated. Moreover, if ﬁgural search is indeed comparable to visual search, this unambiguous condition should yield reaction times similar to those in Experiment 3. So, both the ambiguous and the unambiguous conﬁguration conditions of Experiment 4 are to be compared with the results of Experiment 3, rather than with each other. 5. Experiment 4: ﬁgural search for an unknown target color 5.1. Method 5.1.1. Participants Sixteen participants (13 female, 3 male) were recruited, either in return for course credit or for a small payment. All had corrected or corrected-to-normal vision, and they were unaware of the purpose of the experiment. 5.1.2. Apparatus The apparatus was the same as used in Experiment 3. 5.1.3. Stimuli The ambiguous stimuli used in this experiment were the same as those used in Experiment 3. The unambiguous stimuli were closely related to the ambiguous forms (see Fig. 7 for examples of displays with unambiguous ﬁgure–ground relationships). The individual shapes were identical to the shapes in the ambiguous displays. The background rectangle was again included, and had the same dimensions as the ambiguous stimuli (3.7 · 12.4). However, for the unambiguous displays the individual shapes were vertically displaced to enable them to be unambiguously coded as ÔﬁguresÕ (being small closed shapes against a much larger background). The screen was subdivided into ﬁve horizontal bands (2.3 wide), and each of the objects was randomly positioned in one of the bands. The maximum vertical overlap between two neighboring objects was 1.4. The maximum height of an unambiguous display was therefore 14.4, the minimum height 10. There was no horizontal overlap, because the objects had horizontal positions similar to those in the ambiguous displays. For each of the wide base displays the conﬁguration (dispersion) of the objects was chosen randomly, and the same conﬁguration was then used for the wide top displays. Again two color versions of the stimuli were used, and both these color versions were also inverted. 5.1.4. Procedure The participants were tested individually in a session that lasted approximately 40 min. They were ﬁrst given the instructions for their task, and then they performed a practice block of 16 trials. When the participants felt at ease with the task, they would start the experimental blocks, otherwise they would get another practice block of 16 trials. A trial consisted of a ﬁxation cross presented in the center of the screen for 1000 ms. After the ﬁxation cross disappeared, the stimulus display was presented until the participant responded or 30,000 ms had passed. After the response, there was an inter-trial interval of 1000 ms, before a new trial would start. The task was to look for a vertically symmetric object. It could be either black or white, and it was each color equally often. If the target was present participants had to push the present key on the keyboard as quickly as possible. If it was absent, they had to push the absent key. Responses were made using the ÔZÕ and ÔMÕ keys, with the key used for the response counter-balanced over participants. Ambiguous and unambiguous displays were intermixed. 5.1.5. Design There were four within-subject factors: conﬁguration (ambiguous, unambiguous), top–bottom polarity (wide base, wide top), target presence (symmetric target present, symmetric target absent), and the color of the stimuli (black, white). All the factors were fully crossed. In total, there were 512 trials. There were eight blocks, containing 64 trials each. Between every block there was a self-paced break for the participant. Pilot work indicated high error rates, so we did not retake any of the error trials, to prevent demotivation of the participants. J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 2787 Fig. 7. Examples of displays with unambiguous ﬁgure–ground relations used in Experiment 4: (A) unambiguous black targets and (B) unambiguous white targets. 5.2. Results The results are shown in Fig. 8. Responses greater or less than 2.0 SDs from the cell mean (conﬁguration · top–bottom polarity · target presence · color) for each participant were rejected. This resulted in rejection of 4.3% of the data. The error analysis was performed on all of the remaining trials, whereas the reaction time analysis only used the correct trials. A four-way repeated measures ANOVA on the reaction time data yielded a host of signiﬁcant main eﬀects and interactions. All four main eﬀects were signiﬁcant: conﬁguration F(1,15) = 31.0, p < 0.001, target presence F(1,15) = 63.0, p < 0.001, top–bottom polarity F(1,15) = 11.5, p < 0.004 and color F(1,15) = 14.5, p < 0.002. Moreover, there were signiﬁcant interactions between conﬁguration and target presence F(1,15) = 15.8, p < 0.002, conﬁguration and top–bottom polarity F(1,15) = 7.7, p < 0.015, conﬁguration and color F(1,15) = 4.8, p < 0.045, and a three way interaction between conﬁguration, target presence and color F(1,15) = 8.0, p < 0.015. To analyze the data further, we separated the results for ambiguous and unambiguous conﬁgurations. For the unambiguous conﬁguration, there was only a signiﬁcant eﬀect of target presence F(1,15) = 82.4, 2788 J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 eﬀects of target presence, F(1,15) = 23.2, p < 0.001 and top–bottom polarity, F(1,15) = 9.8, p < 0.005 and an interaction between the two, F(1,15) = 9.5, p < 0.01. The error rate was greater on present than absent trials, and for targets with a wide top relative to a wide base. There was no eﬀect of color and no reliable interactions involving this factor after Bonferroni-correction. 5.3. Discussion Fig. 8. Results of Experiment 4 as a function of color, target presence and the ambiguity of ﬁgure–ground relations. Top panel: reaction times. Bottom panel: error rates. The error bars indicate standard errors. p < 0.001. None of the other main eﬀects or interactions approached signiﬁcance (pÕs > 0.15). For the ambiguous conﬁguration, there were signiﬁcant main eﬀects of target presence F(1,15) = 82.4, p < 0.001, top–bottom polarity F(1,15) = 12.4, p < 0.004 and color F(1,15) = 10.4, p < 0.006. No interactions were reliable with a Bonferroni corrected a of 0.025. Reaction times were faster on present than absent trials, for white relative to black shapes, and for targets with a wide base relative to targets with a wide top. A four-way repeated measures ANOVA on the error data resulted in signiﬁcant main eﬀects of conﬁguration, F(1,15) = 23.7, p < 0.001, target presence, F(1,15) = 26.0, p < 0.001, top–bottom polarity, F(1,15) = 8.8, p < 0.01. Moreover, there were signiﬁcant interactions between conﬁguration and target presence, F(1,15) = 16.7, p < 0.001, between conﬁguration and top–bottom polarity, F(1,15) = 8.9, p < 0.01, and target presence and top–bottom polarity F(1,15) = 7.7, p < 0.015. There was also a three-way interaction between conﬁguration, target presence and top–bottom polarity, F(1,15) = 11.2, p < 0.005. Splitting the analysis along the conﬁguration dimension, for unambiguous stimuli there was only a signiﬁcant eﬀect of target presence, F(1,15) = 7.5, p < 0.02. For ambiguous stimuli there were signiﬁcant main There are several interesting aspects about the results. First, we found a substantial eﬀect of top–bottom polarity in the ambiguous displays. However, this the eﬀect of polarity on reaction times was only apparent on present trials (t(15) = 3.28, p < 0.006) and not on absent trials (t(15) = 0.90, p < 0.38). This is understandable. In order to establish that there is no target present, the participants will have to search through all the items in the display, with both ﬁgure–ground assignments. Consequently, any eﬀect of top–bottom polarity on initial ﬁgure–ground assignment would be minimized. In contrast, target present trials should be inﬂuenced by the initial ﬁgure–ground assignment, which in turn is modulated by the top–bottom polarity of the shapes. Targets would be found faster when they are revealed by the ﬁrst ﬁgure–ground assignment rather than following a second assignment process. The second result of interest is that, for exactly the same (ambiguous) displays, the polarity eﬀect on present trials was larger in Experiment 4 than Experiment 3 (one-sided t(30) = 1.9, p < 0.035). Since the ambiguous displays were generated in exactly the same way for both experiments, the enhanced eﬀect probably reﬂects the fact that participants could use top-down knowledge about the targetÕs color in Experiment 3, but not Experiment 4, although diﬀerences in response strategies between the experiments can not be totally excluded. In Experiment 3, foreknowledge of the targetÕs color should have helped to bias ﬁgure–ground assignment, counter-acting any bias from the top–bottom polarity of the shapes. When the color bias was reduced (in Experiment 4), a stronger eﬀect of top–bottom polarity emerged. Thirdly, we note that RTs for the unambiguous displays are in the same ball park as those for the ambiguous displays in Experiment 3 (t(30) = 1.32, p < 0.196). This is consistent with our suggestion that ﬁgural search is comparable to visual search. The combination of results from Experiments 3 and 4 also discounts two alternative explanations that might be suggested for the observed eﬀects of top–bottom polarity in ﬁgural search. The ﬁrst of these is that the results reﬂect an eﬀect of top–bottom polarity on symmetry detection per se (i.e. that it may simply be the case that it is easier to detect a symmetrical target when it has a relatively wide base J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 and a narrow top than vice versa). Indeed, in another paper (Hulleman & Humphreys, 2004) we have shown that there actually is at most a 10 ms advantage for the detection of vertical symmetry in a single wide-base stimulus relative to a wide-top stimulus. A much larger advantage would be necessary to explain the 100 ms difference we found in Experiment 3. Moreover, even if we would assume that the reaction time advantage for symmetry detection in wide-base stimuli could be larger in the ambiguous displays we have used in Experiments 3 and 4, the results of Experiments 3 and 4 cannot be brought into register using diﬀerences in symmetry detection alone. An explanation purely based on diﬀerences in symmetry detection would predict a smaller effect of top–bottom polarity in Experiment 4. In Experiment 3, the participants know the target color and will therefore only select the items in that color (the fact that the RTs in the unambiguous condition of Experiment 4 are on a par with the RTs in Experiment 3 suggests that this is the case). Hence, in Experiment 3, the reaction time diﬀerence between wide base and wide top targets should be maximal, because the reaction times in either condition will solely be based on the ÔquickÕ wide base items and the ÔslowÕ wide top items, respectively. In Experiment 4, the participants do not know the target color beforehand. Hence, they may frequently select the ÔwrongÕ target color. This will reduce the reaction time diﬀerence between the widebase and the wide-top condition, because the reaction times in both conditions will be based on a mix of the ÔslowÕ wide-top and ÔquickÕ wide-base items. However we observed exactly the opposite: the diﬀerence between the wide-top and wide-base conditions increased in the target present condition of Experiment 4 compared with Experiment 3. Second, the eﬀect of top–bottom polarity cannot be attributed to a speed–accuracy trade-oﬀ. It is true that there were more errors on target present than on target absent trials, particularly for the displays with ambiguous ﬁgure–ground relations in Experiment 4. This pattern, with more errors on target present than on target absent trials, is quite common in visual search tasks. Typically the result is taken to reﬂect the early termination of search on target present trials, prior to the target being detected (e.g. Chun & Wolfe, 1996). In our experiments this early termination of search is most likely to have occurred when the participants assigned the region that contained the target to be ground. The resulting ﬁgural area would contain four asymmetric search items. After ﬁnishing the search of these four items, the participants probably were tempted to terminate their search, rather than attempt to reverse the ﬁgure–ground assignment and search the four remaining items. This mind set would have been encouraged by the nonambiguous stimuli, that eﬀectively contained only four search items. 2789 The result does not detract from the eﬀects of top– bottom polarity on performance however. The critical aspect of the data is that there are more errors for wide top items than for wide base items. Here, the error pattern follows the RT pattern exactly. There are less errors in the quick wide base condition, irrespective of whether the general error level is higher, like on present trials, or lower, like on absent trials. We suggest then that participants conduct a search task based on a representation parsed into ﬁgure and ground regions, and that this regional assignment is inﬂuenced by the top–bottom polarity of the shapes. It could be argued that the RTs in Experiment 4 are too long to reﬂect the inﬂuences of ﬁgure–ground assignment. However, visual search for symmetry is very slow (see also Olivers & Van der Helm, 1998). So, even when the appropriate ﬁgure– ground assignment takes place, to enable the target to be selected, RTs will remain long and likely aﬀected by the number of ﬁgural regions in the display. In addition, there will be large costs on performance if the ÔwrongÕ color is initially interpreted as deﬁning the ﬁgural regions (e.g., for wide top targets, where the ﬁgural assignment might ﬁrst be given to the wide base regions). 6. General discussion We have reported four experiments into the eﬀects of the top–bottom polarity of spatial regions on ﬁgure– ground organization. In Experiment 1, wide base regions were reported as ﬁgure, in preference to wide top regions, in a substantial majority of participants. In Experiment 2, a VSTM-task suggested that participants prefer a wide base shape as a ﬁgure. Experiments 3 and 4 provided additional evidence for the inﬂuence of top–bottom polarity on ﬁgure–ground assignment. In Experiment 3 participants were set to detect a symmetric target in a given color, when there were ambiguous ﬁgure–ground relations between the parts of the display. Despite participants being set to code the display on the basis of color, we found eﬀects of the top– bottom polarity of the stimuli on performance. It was harder to detect a target with a wide top relative to a target with a wide base. In Experiment 4, we examined the same task with the same ambiguous stimuli. This time however, participants were not given information about the targetÕs color beforehand. Under these conditions the eﬀects of top–bottom polarity on performance were far larger than in Experiment 3, where any biasing eﬀect of polarity on ﬁgure–ground assignment could be counter-acted by foreknowledge of the targetÕs color. This interaction between polarity and foreknowledge of the targetÕs color (in Experiment 4 vs Experiment 3), is difﬁcult to explain if polarity only inﬂuenced symmetry detection. However, the results from Experiments 3 2790 J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 and 4 ﬁt with the proposal that top–bottom polarity inﬂuences ﬁgure–ground assignment. In sum, our results converge in suggesting that top– bottom polarity is one factor which, along with others, contributes to ﬁgure–ground coding. Indeed, the combined inﬂuence of factors is demonstrated by the contrasting strength of the polarity eﬀects in Experiments 3 and 4. In Experiment 3, the inﬂuence of top–bottom polarity was moderated by top-down parsing by color into ﬁgure and ground. Prior studies of ﬁgure–ground coding have not indicated that top–bottom polarity is a critical factor, though factors such as object familiarity (ÔdenotivityÕ in Peterson & GibsonÕs, 1994 terms) have been shown to play a part and object familiarity is likely reduced by inversion (e.g., see Jolicoeur, 1985). It might be argued that the wide base stimuli we used in our experiments look like the silhouettes of evergreen trees. Since these are familiar shapes, this might have driven ﬁgure–ground assignment, rather than top–bottom polarity. However, there are several diﬀerence between the wide base regions and evergreen trees. First, the silhouettes of evergreen trees tend to be vertically symmetrical, due to their exposure to gravity. All wide base regions in Experiments 1 and 2, and most wide base regions in Experiments 3 and 4 were asymmetrical. Second, the silhouette of an evergeen tree tends to be a smooth curve, unlike the serrated envelopes of the wide base regions in our experiments. Third, because evergreen trees are trees, they actually have a narrow base: their stem sticks out from underneath the branches. Moreover, it could be argued that the wide top shapes look like the tornade shapes that are familiar from weather programs and disaster movies. Clearly there was no advantage for these forms. Another argument against familiarity as an explanation of our results are the stimuli used by Peterson and Gibson (1994). Their displays contained regions with easily recognisable shapes that had to compete against regions with no reasonable object interpretation at all. Even under those more ideal circumstances Peterson and Gibson (1994) reported for the condition where both the high and the low denotative regions were asymmetric (comparable to our Experiment 2) only a 61% preference for the high denotative regions. It seems inappropriate to call either the wide base or the wide top stimuli in Experiment 2 highly denotative, suggesting also that the inﬂuence of familiarity on the outcome of Experiment 2 was minimal. As we have noted, Vecera et al. (2002) proposed that there is a tendency to code the lower region of a stimulus as ﬁgure and the upper as ground, when a display segments into distinct upper and lower regions. From this it might be argued that, here, the bottom part of the shapes was strongly weighted in ﬁgure–ground organization, so that there was a bias to code a stimulus as ﬁgure if it had a wide/stable bottom part. So, rather than a new ﬁgure–ground cue, the top–bottom polarity should be considered as an extension of the lower region cue. However, in our case, whole shapes (top, as well as bottom parts) were coded as ÔﬁgureÕ, so there was no simple parsing of the display into bottom and top halves. Indeed, unlike Vecera et al. (2002), the bottom half of the displays contained ground as well as ﬁgural elements. Furthermore, the diﬀerence in the distribution of the area between wide base and wide top stimuli was relatively minor. In Experiment 2, the centres of mass were, on average, respectively some 6% of the height of the stimuli below and above the midline of the displays. So, most of the time, the centres of mass of the wide base stimuli were in the ﬁfth horizontal strip from the bottom, just under the midline of the display. Describing this situation as lower region probably stretches beyond the point where it is useful to use the concept. This is especially the case the stimuli with 16 horizontal strips (used in the other experiments), where the centres of mass fell even closer to the middle. Now, when given a choice as to whether a Ôwide baseÕ or a Ôwide topÕ stimulus is upright, independent observers routinely choose the Ôwide baseÕ stimulus. Thus the eﬀect of top–bottom polarity on ﬁgure–ground assignment may well reﬂect a more general bias in visual coding, favoring objects that appear stable when aligned with the gravitational upright. It is important to point out that our argument for the eﬀects of top–bottom polarity on ﬁgure–ground coding comes from several sources. First, we found the bias when we employed an explicit reporting task. Second, we observed the same bias when we employed a VSTM-task. Third, we also found the eﬀect of top–bottom polarity in a novel task that depends on ﬁgure– ground assignment, but is not directly contingent on introspective factors. This convergence of results not only supports the conclusion that top–bottom polarity is a ﬁgure–ground cue, it also suggests that ﬁgural search has some credence as a method in ﬁgure–ground research. We propose that ﬁgural search is a welcome addition to the instruments at the disposal of researchers interested in ﬁgure–ground assignment. Acknowledgment This work was supported by the UK Medical Research Council. References Baylis, G. C., & Driver, J. (2001). Perception of symmetry and repetition within and across visual shapes: Part-descriptions and object-based attention. Visual Cognition, 8, 163–196. J. Hulleman, G.W. Humphreys / Vision Research 44 (2004) 2779–2791 Chun, M. M., & Wolfe, J. M. (1996). Just say no: How are visual searches terminated when there is no target present? Cognitive Psychology, 30, 39–78. Driver, J., & Baylis, G. C. (1996). Edge-assignment and ﬁgure–ground segmentation in short-term visual matching. Cognitive Psychology, 31, 248–306. Hulleman, J., & Humphreys, G. W. (2004). Is there an assignment of top and bottom during symmetry perception? Perception, 33, 615–620. Jolicoeur, P. (1985). The time to name disoriented natural objects. Memory and Cognition, 13, 289–303. Metzger, W. (1936). Gesetze des Sehens (Laws of seeing). Frankfurt am Main: W. Kramer & Co. Olivers, C. N. L., & Van der Helm, P. A. (1998). Symmetry and selective attention: A dissociation between eﬀortless perception and serial search. Perception and Psychophysics, 60, 1101–1116. 2791 Palmer, S. E. (1999). Vision Science: Photons to Phenomenology. Cambridge, MA: MIT Press. Peterson, M. A., & Gibson, B. S. (1994). Must ﬁgure–ground organization precede object recognition––an assumption in peril. Psychological Science, 5, 253–259. Strüber, D., & Stadler, M. (1999). Diﬀerences in top-down inﬂuences on the reversal rate of diﬀerent categories of reversible ﬁgures. Perception, 28, 1185–1196. Treisman, A., & Souther, J. (1985). Search asymmetry––a diagnostic for preattentive processing of separable features. Journal of Experimental Psychology: General, 114, 285–310. Vecera, S. P., Vogel, E. K., & Woodman, G. F. (2002). Lower region: a new cue for ﬁgure–ground assignment. Journal of Experimental Psychology: General, 131, 194–205. Wolfe, J. M. (2001). Asymmetries in visual search: an introduction. Perception and Psychophysics, 63, 381–389.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project