One aspect of perceptual organization is figure-ground segregation. If a particular visual stimulus is blobs of contours (which it is at the retinal level), the figure is an integrated group of contours (Coren et al. 1993), and the ground is the background against which it stands.

The phenomenon of figure-ground segregation is of interest in the field of perception research because of its importance in everyday life and its evolutionary survival value. A simple example: we would not be able to pick berries from among the leaves if we could not designate what in the scene is the figure (i.e., the berries), and what the background (i.e., the leaves).

When a figure is perceived, often not all of its contours are actually detected at the retina– some of them are subjective contours, which are not physically present at the retina, but are the product of intelligent perception. This demonstrates that the phenomenon of figure-ground segregation is not a purely bottom-up process (i.e., it is not simply data-driven). Rather, it is bottom-up (data-driven) as well as top-down (conceptually-driven).

The interaction between bottom-up and top-down processing becomes especially evident in the case of form illusions in, for example, reversible figures such as the famous Rubin (1915, 1958) face-vase figure (Figure 1). Depending on what we designate to be the figure and what the ground, we perceive the same stimulus as two different things: two faces in profile facing each other (against a white background) or a vase (against a black background).

Figure 1 . Rubin ambiguous face-vase figure.

Recently, researchers have disputed over whether a cognitive theory of the perception of illusionary figures (whose ambiguity is based on the figure-ground segregation) is an appropriate explanatory theory. The cognitive theory was initially proposed by Gregory (1972) and further developed by Rock and Anson (1979). It holds that subjective contours are due to top-down processing, and that they are the product of the problem-solving efforts of the subject. This theory was criticized by Spillmann and Dresp (1995), who dispute the separation of top-down and bottom-up processing, and argue that an integrative approach combining local feature detection and global strategies of perceptual organization provides a more complete explanation of illusory figure perception. They further point out that, for a theory of perception to have explanatory power, it needs to consider the neurophysiology underlying the perceptual mechanism, which means that such a theory cannot deal solely with top-down processing.

In response to these arguments, Parks (2001) defends the Gregory-Rock theory by saying that Spillmann and Dresp (1995) have misrepresented it. Parks argues that, though Gregory-Rock’s explanation of illusory figure perception invokes the idea that prior experiences (i.e., higher cognitive factors, such as learning and memory) guide such perceptual phenomena, this idea is not central to the theory. The central claim is that perception is a problem-solving activity, where the stimulus is the problem and the various perceptual interpretations are possible solutions. Parks suggests that it is an activity heavily influenced by the “degree of coincidence” in any possible perceptual solution.

In response, Dresp and Spillmann (2001) asserted that the Gregory-Rock theoretical framework “is not appropriate for suggesting candidate mechanisms of brain-behavior functions that could underlie the phenomenal emergence of [illusory] figures” because the theory cannot predict subjects’ responses in tasks presenting them with ambiguous figure-ground stimuli.

Furthermore, research findings indicate different cues for figure-ground assignment. If real, these findings would provide the cognitive theory with the required explanatory power, since they would predict subjects’ responses. An example is Vecera et al.’s (2002) report that “regions in the lower portion of the stimulus array appear more figure-like than regions in the upper portion of the display” (Figure 2). Another cue is the size of areas and shapes on the display. Smaller areas or shapes are more likely to be perceived as figures (Coren et al. 1993). Also, evidence indicates that areas of the display that are designated as figures are interpreted as having a shape; whereas, what is taken to be the ground is not interpreted as having a shape (e.g., Baylis and Cale 2001). Other factors for the figure-ground designation include perceived spatial organization, luminance contrast, and sector angle (e.g., Shank and Walker 1989).

Figure 2. Examples of lower-region cue to figure-ground assignment: both A and B are horizontally symmetrical, and have black and white regions of equal size. Despite that, and even though the two figures are actually color-reversed versions of each other, people show preference for seeing the black region as the figure in A and the white region as the figure in B due to the lower-region cue.

In light of this, four important concepts from the field of perception are related when it comes to figure-ground reversible displays: figure-ground ambiguity, bottom-up versus top-down processing, local versus global perception, and semantic priming. These concepts and their interaction have not been studies together yet, and it is unclear how they operate together during performance of perceptual tasks. A fuller, more detailed investigation of the relations among these perceptual factors is needed. The first two of these factors are discussed above; the second two factors, below.

The question of local versus global perception has to do with aspects of the display. The local aspects are the detailed aspects of a figure; whereas, the global ones are larger-scale, which determine the perception of the display as a whole. Global precedence is at work in most cases: people attend to global features of the stimulus before they attend to local ones (Coren et al. 1993). This is tied in with the levels of processing. Most theories that emphasize data-driven processing also emphasize the importance of local features. However, evidence for global precedence brings in higher levels of cognition. With respect to figure-ground segregation, this phenomenon seems to be connected to global aspects of perception.

Finally, semantic priming inevitably affects the debate on conceptually-driven versus data-driven processing and global versus local aspects of a stimulus, because, if a perceptual phenomenon is (at least theoretically) associated with top-down processing and global perception, then subjects should show susceptibility to semantic priming with respect to that phenomenon. Studies have shown that semantic priming indeed readily occurs with figure-ground ambiguous stimuli. For example, Davis et al. (1990) report that appropriate semantic priming facilitates figure-ground organization, and Girgus et al. (1977) found that, if they told their subjects that the stimuli they were shown were reversible figures, for the same allowed period of viewing (three minutes), all of the participants were able to reverse the figures easily and frequently (as opposed to half of them, when they were not informed of the ambiguity).

According to Vecera and O’Reilly (2000), figure-ground segregation occurs before object recognition. This means that, when we detect an array of contour blobs at the retina and this information gets processed by the visual system, we try to designate the contours that form an object. This process precedes the recognition of the object.

This study investigated these four perceptual issues and the interactions among them by investigating the effects of form (i.e., figure-ground) ambiguity and the availability of cues (i.e., semantic priming) on the perception of the intended meaningful aspect of ambiguous stimuli. We also investigated how the presence of figure-ground segregation alternatives influences object perception. The manipulated variables were: nature of the visual stimulus (blob versus word; Figure 3), and semantic priming (present or absent). Reaction time to correctly describe or name the stimulus was measured.

Figure 3. Word and blob stimuli used in this experiment.

With respect to figure-ground ambiguity, we hypothesized that subjects would perform significantly faster on describing or naming an unambiguous familiar stimulus (a word) than an ambiguous unfamiliar one (a blob).

With respect to top-down versus bottom-up processing, we hypothesized that, with unfamiliar stimuli, bottom-up processing would occur first, before top-down processing, and, therefore, on the first figure-ground ambiguous (blob) trial the subjects would have a significantly longer reaction time than on subsequent trials. Furthermore, reaction time would decrease with trials because each subsequent trial would be cued so the subjects would be semantically primed as to what to expect.

Finally, with respect to semantic priming, we predicted that subjects in group A would show higher reaction time results than subjects in group B, because the former were going to be cued by the words they were to view prior to viewing the ambiguous stimuli.

Apparatus and materials
Seven four-letter word stimuli and seven blob (ambiguous) stimuli were used. The words were typed in purple on a large white background, while the blobs were comprised of purple ink stains with no intended shape on the same white background. The portions of white among the purple blobs also formed four-letter words. The two types of stimuli were of approximately equal size on the display (Figure 3), and each blob had a corresponding word such that at least two letters were the same. In each group (words vs. blobs), six were nouns and one was a verb. They were all familiar, frequently occurring four-letter English words.

The ambiguity of the blob patterns comes from a figure-background designation. When the colorful patches are perceived as the figure, the pattern is meaningless. This is what happens as a result of bottom-up processing, because the white letters blend with the surrounding white to form the "natural" background. However, when the colorful patches are viewed as the background, then the white patches are the figure and they pop up as letters forming meaningful words.

Design and procedure
This was a 2 x 2 mixed factorial design using combined assignment. The two independent variables were the nature of the stimuli and the order of presentation, and the dependent variable was reaction time for meaningful perception.

The nature of the stimulus is operationally defined in terms of the figure-ground ambiguity in the stimulus. This variable has two levels: word (figure-ground ambiguity is absent) or blob (figure-ground ambiguity is present). This was the repeated measures variable, and participants were treated with both levels of this variable.

Order of presentation is operationally defined as the sequence of stimuli of different nature. The variable has two levels: words-blobs or blobs-words. This was the independent group’s variable. Participants were randomly assigned to the two levels of this variable (group A: words-blobs, or group B: blobs-words). Subjects in group A were first presented with a sequence of the seven word stimuli, followed by a corresponding sequence of the seven blob stimuli, while subjects in group B were first presented with the blob stimuli, followed by the word stimuli.

Furthermore, to avoid order effects, partial counterbalancing was adopted. Complete counterbalancing was not possible due to the limited number of subjects. Groups A and B were thus further subdivided each into six groups where the order of stimulus presentation was varied. The specific order-of-presentation sets for each subgroup are presented in Table 1. The counterbalancing concerned only one of the two types of stimuli, the words, and the order of presentation for the blobs was then accorded with the order for the words.

Table 1 . Stimuli According to their Order of Presentation in the Different Experimental Groups.

The reaction time for meaningful perception is operationally defined as the time between presentation of the stimulus and the subject’s report of perceiving the meaningful (word) aspect of the stimulus. This was measured in seconds using a stopwatch.

Each participant was tested individually. Participants were presented with the stimuli on a computer monitor, which was at a standard distance from their eyes (approx. 50 cm). There was only one stimulus at a time: a word or a blob. The participant was asked to describe or name what they saw. The time between the beginning of the viewing and the report was measured. In cases where the participant failed to report perceiving the meaningful aspect of an ambiguous stimulus, s/he was given up to a minute (this was considered to be the ceiling), and then the current stimulus was removed and a new one presented. There were only short pauses between trials (up to 5 seconds).

Table 2 . Raw Data Reaction Times (in seconds) Taken by Individual Subjects to Correctly Name the Word and Blob Stimuli. Sorted by order of presentation. (Click to view full table)

Table 3. Raw Data Reaction Times (in seconds) Taken by Individual Subjects to Correctly Name the Word and Blob Stimuli. Sorted by word or blob. (Click to view full table)

Figure 4 . Average reaction times on the blob stimuli for subjects in groups A and B. This graph examines the effects of order of presentation. Actual reaction times (in seconds) are listed in the table at the bottom of the graph, with standard deviations (in seconds). (Click to view enlarged table)

Figure 5. Average reaction times on the word stimuli for subjects in groups A and B. This graph examines the effects of order of presentation. Actual reaction times (in seconds) are listed in the table at the bottom of the graph, with standard deviations (in seconds). (Click to view enlarged table)

Figure 6. Average reaction times on the blob and word stimuli for all. This graph examines the effects of order of presentation. Actual reaction times (in seconds) are listed in the table at the bottom of the graph, with standard deviations (in seconds). (Click to view enlarged table)

From Table 4 and Figure 12, one can see that, on average, subjects’ correct naming responses were much faster and much less variable on the word stimuli than on the blobs. Furthermore, Figure 6 shows a noticeable descending trend in the reaction times and in the variability (standard deviation) of the reaction times with increasing order of the stimulus. This trend is much more prominent for the blob stimuli than for the words. From Figure 4, one can see that this “blob reaction-time descending trend” occurs with the results for group A alone, but not with the results for group B, which seem much more irregular in their pattern with respect to order of the stimulus and have much more variability. Nonetheless, both groups show a considerable decrease in reaction time from the first to the second blob stimulus (Figure 4), and the results for group B are consistently higher than the ones for group A. On average, as seen in Table 4 and Figure 12, the mean reaction time for the blob stimuli was 10.2 seconds for group A versus 21.4 seconds for group B. On the other hand, as can be seen in Figure 5, the results for the word stimuli are similar for the two groups and show little variability. In fact, on average, the mean reaction time for the word stimuli was 1.2 seconds for both groups (Table 4).

Table 4. Mean Reaction Times and Standard Deviations.

Figure 12. Mean reaction times and standard deviations for blobs and words.

However, despite these noticeable trends in the results, Figure 7 and Table 5 show that most of the trends are not significant. Figure 7 illustrates that, as expected, the average differences in performance between groups A and B on the word stimuli are not significant: all Hedge’s g values are |g| < 1 STD unit. Moreover, contrary to what was expected, the graph also shows that the average differences in performance between groups A and B on the blob stimuli are not significant: all Hedge’s g values (but one) are |g| < 1 STD unit.

Table 5. Hedge’s g Values Showing Overall Trends in the Results.

Figure 7. Hedge’s g values on the blob and word stimuli for all subjects. This graph examines the effects of order of presentation. (Click to view enlarged table)

Figure 8. Average reaction times on the blob stimuli for subjects in groups A and B. This graph examines the effects of word difficulty. Actual reaction times (in seconds) are listed in the table at the bottom of the graph, with standard deviations (in seconds). (Click to view enlarged table)

Figure 9. Average reaction times on the word stimuli for subjects in groups A and B. This graph examines the effects of word difficulty. Actual reaction times (in seconds) are listed in the table at the bottom of the graph, with standard deviations (in seconds). (Click to view enlarged table)

Figure 10. Average reaction times on the blob and word stimuli for all subjects. This graph examines the effects of word difficulty. Actual reaction times (in seconds) are listed in the table at the bottom of the graph, with standard deviations (in seconds). (Click to view enlarged table)

Of special concern to this investigation was that the performance on the first blob stimulus be significantly different in the two groups, since the subjects in group A were semantically cued while those in group B were not. Contrary to that prediction the Hedge values indicate that there is only a negligible difference between the two groups (g = 0.24 STD units). The only Hedge’s g value to note is the one for the fourth blob stimulus, g = 1.12, which indicates the predicted improvement in reaction time performance when going from group B to group A. However, by itself, this value is insignificant, and is probably a side effect of the limited number of subjects and/or the varying difficulty level of the stimuli in combination with the lack of complete counterbalancing to guard against this unfortunate variability.

Table 5 indicates that the expected change in reaction time between the first blob stimulus and the second blob stimulus was not significant, even though the Hedge’s g values indicate the predicted direction of improvement when going from the first to the second stimulus and the magnitude of this improvement was uniform for all subjects, independent of which group they were in. Similar results were found for the word stimuli, even though they exhibit greater variability. As expected, there was no difference in the average performance of subjects from A and B on the word stimuli (g = 0.0), and there was a difference for the blob stimuli in the predicted direction; however, this difference was not significant (g = 0.55 < 1 STD unit).

The only trend that seems significant is the decrease in reaction time between the blob and the word stimuli when going from the first presented blob to the first presented word: for group A g = – 1.11, for group B g = – 1.69, and overall: g = – 1.42. Change on average from blobs to words for group A: g = – 0.80; for group B: g = – 1.19; and overall: g = – 0.98.

In terms of the individual stimuli, Figures 8 and 12 show that there was indeed a considerable variability in the difficulty level of the blob stimuli, with ‘seek’ being the hardest, receiving an average reaction time of 21.5 seconds for identification, and ‘loop’ being the easiest, receiving an average reaction time of 9.0 seconds for identification. Furthermore, there were differences between the two groups: both groups found items like ‘seek’ difficult (19.2 sec for group A and 23.9 sec for group B), while for other items, like ‘boat’, one group performed much worse than the other (6.2 sec for group A and 31.4 sec for group B). There also seems to be some variability in the difficulty level of the word stimuli, with ‘bulb’ receiving highest reaction times in both groups (Figure 9).

Figure 11. Hedge’s g values for word and blob stimuli. This graph examines the effects of word difficulty. (Click to view enlarged table)

The fact that, on average, the reaction time for the different word stimuli was the same (apart from the word ‘bulb’) supports the idea of global precedence. If the word identification depended on the processing of local features alone, then there would have been more variability since the features are different in the different words. Were the blob stimuli of an equal difficulty level, we would expect them to show little variability in reaction time on average since their identification also depends mainly on global perception.

Interestingly, even though all subjects were told that this was an ambiguity perception study, subjects were often not able to identify the word in a blob stimulus for as long as 1 minute. This seems to contradict the findings of Girgus et al. (1977) that knowledge of ambiguity greatly facilitates disambiguation. However, in the present study, the effect of knowledge of ambiguity was not investigated, so no conclusions can be drawn in this respect. Furthermore, it is possible that the stimuli were altogether too difficult to demonstrate the effect of this variable.

Observations from this study seem to confirm the findings of Vecera and O’Reilly (2000) that figure-ground segregation occurs before object recognition. Even though, on many occasions, for periods of several seconds, subjects were not able to describe the ambiguous stimulus, when they finally did (describing it as ‘patches of purple ink’ or ‘purple birds’ or ‘map’), their description identified the purple blobs as the figure, even though the shape of this figure was hard for them to describe. This shows that they had already designated the blobs to be the figure before they could recognize the object depicted by this figure.

As for the predictions that the first figure-ground ambiguous (blob) trial subjects would have a significantly longer reaction time than for subsequent trials and that reaction time would decrease with trials, our findings somewhat confirm the first part of the hypothesis, but are absolutely inconclusive about the second part. The results were insignificant, so interpretation would only be speculative and no conclusions should be drawn.

Limitations

If the hypothesis is wrong, then our results indicate precisely what they should. However, if the hypothesis can be demonstrated to hold, then there are a number of reasons why our results did not accomplish this. The limitations of this study include the small sample size, the special population (of psychology perception students) from which this sample was drawn, the nature of the stimuli, and the cultural factor in perceptual organization, which was not taken into account in the present study.

First, the small sample size and the special population from which it was drawn might have had a number of effects on the results. Small sample sizes allow greater manifestation of individual differences, such as imaginatively, reading, cultural and language background, gender, age, and up-bringing. It has been reported that individual differences are a considerable factor with respect to figure-ground perception (e.g., Forsyth and Huber 1976). For example, studies show that males perform better on figure-ground perception ability tests (e.g. Davis 1995), and that figure-ground discrimination is poorer in older individuals (Roper et al. 2001). Furthermore, “word-reading speed and naming speed of colors and pictures continue to increase into mature adulthood” (Van den Bos et al. 2002). There were only two males among the 14 subjects in this study, and all subjects were in the same age group, so it was not possible to investigate the effect of these individual differences (as well as others) on the identification of the ambiguous blob stimuli.

Also, the fact that the sample was drawn was the population of psychology perception students at Simon Fraser University might have had a number of effects, such as specific expectations about the study based on the students’ knowledge of the topic, negative sets and functional fixedness, or careless reading of the instructions.

Second, Ganguli and Broota (1973) have reported that cultural factors play a significant role in the disambiguation of ambiguous figure-ground situations. They have related differences in the performance of subjects with different cultural background on tasks based on figure-ground ambiguity with difference in child-rearing practices. In the present study, cultural differences were not systematically investigated, but they did seem to have an effect: non-native English speakers (i.e., Chinese and Korean subjects) had an average reaction time of 29.8 seconds on the blob stimuli, and native English speakers (i.e., Canadian subjects) had an average reaction time of 5.3 seconds. This considerable difference cannot be attributed to reading abilities, since the two cultural groups performed approximately equally on the word stimuli (with non-native English speakers having an average reading speed of 1.3 seconds per word and native English speakers having an average reading speed of 1.1 seconds per word). Furthermore, the effect of cuing seems to be the same for the two groups, doubling the reaction time when going from the cued group A to the uncued group B (for non-native English speaker, group A showed a mean of 18.8 seconds on the blob stimuli, and group B showed a mean of 40.8 seconds; while for native English speakers, group A showed a mean of 3.7 seconds, and group B showed a mean of 6.9 seconds). This interesting outcome seems to imply that the differences between the two cultural groups is not related to their English reading abilities, or to the effect of semantic priming, but may well be related to their up-bringing and cultural heritage. Such cultural factors can also be used to explain the observed ceiling effects.

Third, another important source of variability may be the nature of the stimuli. The stimuli were especially created for this study, so the possible effects of their particular characteristics could not be avoided. Such characteristics include difficulty level, word choice, and color. This study established that the blob stimuli were indeed of varied difficulty, which certainly could have contaminated the investigation. Furthermore, the choice of words for both the blob and the word stimuli may have been inappropriate. This concerns several issues: word category, concreteness, and word frequency. The words were not of the same category (12 were nouns and two were verbs), and they were not chosen from a reliable source of frequent words, so it is likely that they have different frequency of appearance in everyday speech (definitely, ‘book’ is a more frequently occurring word, especially among students, than ‘seed’ or ‘bulb’). Both of these aspects could have influenced the performance of the subjects. Another issue is that some of the noun words were concrete nouns (e.g., ‘wood’) while others were abstract (e.g., ‘deed’), which also may have had an effect on cognitive processing. Finally, the color of the stimuli (purple) may also have influenced performance, or at least the extent to which semantic cuing had an effect, since we are used to seeing words printed in black. A number of studies have concluded that color has an effect on reading speed. Most studies have dealt with the effect of differently colored backgrounds for reading texts (e.g., Croyle 2000; Wilkins and Lewis 1999), but this effect is due to the luminance contrast and the spectral contrast between the background color and the color of the printed text. Thus, if changing the background color makes a difference, then so should changing the color of the foreground, even if to a lesser degree.

In conclusion, we suggest a number of ideas for future research in the field. First, to test the validity of the hypothesis (which at present is neither rejected nor supported) one could replicate this study with more and/or different blob stimuli and with a larger sample. Second, to investigate the effect of individual differences on reaction time performance for the blob stimuli, one could replicate the study with subjects from different age groups, different gender, and/or different cultural backgrounds. In particular, in light of the above analysis and interpretation, a study of the effect of cultural background on the identification of ambiguous figure-ground (word) stimuli may turn out very fruitful and interesting. Third, to investigate the effect of color, one could replicate the study using the same stimuli but in different foreground colors and/or on differently colored backgrounds. Finally, to investigate the effect of word choice, one could use words of different length, different frequency, different levels of concreteness, and/or different word categories.

Despite the statistically insignificant results of this experiment, this research is important, not only because it gives rise to a number of possible future studies, thus contributing to the field of perception research, but also because it has broader implications, academic as well as practical, for other areas, such as education, sociology and cultural studies. Determining the effect of the foreground color of text, of ambiguous blobs on reading speed, and the identification of ambiguous stimuli can be used in education to facilitate verbal as well as pictorial learning and comprehension. Findings concerning gender, age, individual differences, and culture effects on figure-ground disambiguation can be used in sociology and anthropology in interpreting, for example, social group preferences and practices in child upbringing, art, and situational analysis and behavior.