Visual search in divided areas: Dividers initially interfere with and later facilitate visual search
A common search paradigm requires observers to search for a target among undivided spatial arrays of many items. Yet our visual environment is populated with items that are typically arranged within smaller (subdivided) spatial areas outlined by dividers (e.g., frames). It remains unclear how dividers impact visual search performance. In this study, we manipulated the presence and absence of frames and the number of frames subdividing search displays. Observers searched for a target O among Cs, a typically inefficient search task, and for a target C among Os, a typically efficient search. The results indicated that the presence of divider frames in a search display initially interferes with visual search tasks when targets are quickly detected (i.e., efficient search), leading to early interference; conversely, frames later facilitate visual search in tasks in which targets take longer to detect (i.e., inefficient search), leading to late facilitation. Such interference and facilitation appear only for conditions with a specific number of frames. Relative to previous studies of grouping (due to item proximity or similarity), these findings suggest that frame enclosures of multiple items may induce a grouping effect that influences search performance.
KeywordsVisual search Efficient search Inefficient search Separated area
You are looking for a stapler. When you open a drawer of your desk, you see many stationery products arranged in separate boxes. Finally, you find the stapler in one box in the drawer. As in this scenario, in our daily lives, there are many opportunities to search for some target items within areas segregated by dividers (e.g., books on shelves or food in a refrigerator). How do we look for items in such situations?
A number of previous studies have examined search behaviors using search displays in which items were uniformly distributed over a large search region (e.g., Treisman & Gelade, 1980; Wolfe, 1994). However, we know very little about the nature of search behavior in situations in which multiple search areas are outlined by dividers (e.g., frames). If we wish to understand visual search in our daily lives, it is important to take into account visual search in separated areas.
Recent studies have reported that the global information of a scene (e.g., gist or layout) can influence visual search (e.g., Castelhano & Heaven, 2011; Castelhano & Henderson, 2007; Torralba, Oliva, Castelhano, & Henderson, 2006; Võ & Henderson, 2010). For example, Võ and Henderson suggested that scene representations generated from only a brief glimpse of a whole scene may provide sufficient information to guide gaze during an object search. Therefore, it is possible that the stimuli separating a large visual search area into smaller ones (i.e., dividers) may also affect search behaviors, because the dividers can provide global scene information.
In this study, we examined whether visual search performance is influenced by segregating a visual search area into subsets of smaller search areas. In many previous studies, researchers have argued for two types of visual search: an efficient (i.e., parallel, preattentive) mechanism and an inefficient (i.e., serial, attentive) mechanism (e.g., Treisman & Gelade, 1980; Wolfe, 1994), although the distinction between these two types of searches is somewhat ambiguous (e.g., Wolfe, 1998a). We intended to investigate how such manipulations might affect both efficient and inefficient searches. To accomplish this, we conducted two types of visual search tasks, in which participants searched for either a single C among Os or for a single O among Cs. The former task is an efficient search, whereas the latter is an inefficient search (search asymmetry: see Treisman & Gormican, 1988; Treisman & Souther, 1985). In the two tasks, the items (i.e., target and distractors) that we used were the same shapes; thus, the similarity between the features of the items and the frames was controlled across the two tasks.
We used black square frames, in which either a single large frame surrounded an entire search display or four (or more) smaller frames divided the same display into smaller search areas. A single large frame simply defined the entire visual search area. This was a control condition. The experimental conditions involved four or more smaller frames that divided the larger area. In all frame conditions, the stimulus items were identical; they consisted of white curved lines (i.e., the letters C and O). The black square frames shared no features with the display items. In other words, all features of the frames were completely task-irrelevant. Some previous studies have suggested that entirely irrelevant distractors may interfere with performance in cognitive tasks (e.g., Forster & Lavie, 2008, 2011; see also Folk & Remington, 1998). Therefore, in normal circumstances the presence of divider frames may serve as a distraction, because they introduce unnecessary information. That is, they make the visual environment redundant by adding more to-be-processed information. If the presence of dividing frames functions as a distraction, which essentially could be ignored during a target detection task, visual search performance should decrease with their presence.
Alternatively, dividing frames might function in a positive fashion, by affording guidance in searching for a target due to a facilitation of serial attention allocation. Multiple items within an outlined area can be regarded as a single group (i.e., essentially as one item). Several previous studies have examined the effects of perceptual grouping due to similarities among stimulus items (e.g., Egeth, Virzi, & Garbart, 1984; Kim & Cave, 1999; Niemelä & Saarinen, 2000; Treisman, 1982). For example, Niemelä and Saarinen conducted a visual search task in which observers searched a computer display for one icon with a specific filename among several other types of icons. The researchers compared search performance between a condition in which similar icons (i.e., same file type) were grouped spatially (grouped condition) and a condition in which all of the icons were randomly positioned (ungrouped condition). The results showed that search performance was better in the grouped than in the ungrouped condition. Thus, visual search can be facilitated when objects are spatially grouped. If this is true, a dividing frame that spatially segregates search areas should also improve visual search performance.
A group of 16 undergraduate and graduate students (mean age = 23.4 years) with normal or corrected-to-normal vision participated. They were naive with respect to the purpose of the research. The data from four participants were not collected due to a computer error; the data of the remaining 12 participants were analyzed.
We prepared white C-shaped and O-shaped figures (1.5° × 1.5° of visual angle) as the search items. The C-shaped figure was a circle with a gap on the right side. Two background displays were created: a one-frame and a four-frame display. The one-frame display was a gray background that contained one large square frame (12.5° × 12.5°) at the center of the display. The four-frame display contained four frames, each outlining a smaller spatial area within the larger display (6.0° × 6.0°); each frame was at one of the four quadrants of the larger display. The distance between two contiguous frames in the four-frame display was 0.5°.
We used two set size conditions: four and 16 items. In the four-frame display, one or four items appeared in each frame. The item locations in each frame were selected randomly from an imaginary 3 × 3 grid (the size of each cell was 2.0° × 2.0°). In the one-frame display, the item locations were selected randomly from the same location options used for designating locations in the four-frame display condition. Therefore, the items were distributed identically in both frame conditions (see Fig. 1). Target presence was also manipulated. In the target-present condition, one target, either O or C, was presented with three or 15 Cs or Os, respectively. In the target-absent condition, four or 16 Cs (or Os) were presented.
Presentation of the stimuli and recordings of participant responses were controlled by ViSaGe (Cambridge Research Systems, Cambridge, U.K.). Stimuli were displayed at a resolution of 1,024 × 768 pixels in 24-bit color on a 22-in. monitor.
The participants were seated in front of a computer monitor (viewing distance was 57 cm) in a dim room. On each trial, after a fixation cross (500 ms), a search display appeared and remained until a participant had responded. The participants were instructed to search for a single target (i.e., either an O or a C), identified at the beginning of each experimental block, and to judge whether it was present or absent by pressing the appropriate button as quickly and accurately as possible.
The participants completed two tasks (i.e., visual search for O and for C) in an experimental session of 320 trials, with 40 trials in each of the eight conditions created by the 2 (target present or absent) × 2 (set size 4 or 16) × 2 (one or four frames) factorial design. Task order was counterbalanced across participants. Within each task, the order of trials was randomly determined. Prior to each task, participants completed a practice session of 16 trials, two in each of the eight conditions.
Summary of the mean slope and intercept (in milliseconds) for each condition in visual searches for (a) an O and (b) a C (M ± SE)
(a) Visual-search-for-O task
20.1 ± 2.3
492.6 ± 25.9
40.6 ± 4.6
529.8 ± 32.5
17.4 ± 1.8
521.9 ± 31.3
37.1 ± 4.1
552.6 ± 33.1
(b) Visual-search-for-C task
8.6 ± 1.3
488.9 ± 23.9
18.0 ± 2.5
524.7 ± 22.5
9.2 ± 1.2
506.5 ± 27.9
17.3 ± 2.5
511.2 ± 18.7
We conducted a 2 × 2 repeated measures analysis of variance (ANOVA) for each task type. In each analysis, the effects of the factors Target Presence and Frame on search slopes (i.e., the processing time per item) and intercept (i.e., the fixed early processing time) were evaluated. In the search-for-O task, the slope in the target-absent condition was greater than that in the target-present condition, F(1, 11) = 26.22, p < .001, ηp2 = .70. Furthermore, the presence of a four-frame display significantly reduced the search slope relative to the one-frame condition, F(1, 11) = 9.91, p < .01, ηp2 = .47. The interaction between target presence and frame was not significant, F < 1. In addition, the four frames increased the intercept (i.e., fixed early processing time), F(1, 11) = 7.33, p < .03, ηp2 = .40. Neither the main effect of target presence nor the interaction was significant, Fs < 1.13, ps > .3.
In the search-for-C task, the slope in the target-absent condition was greater than that in the target-present condition, F(1, 11) = 20.51, p < .001, ηp2 = .65. However, in this task neither the main effect of frame nor its interaction with target presence was significant, Fs < 1. In the analysis of the intercepts, neither of the main effects nor the interaction was significant, Fs < 2.92, ps > .1.1
In the inefficient search task, the frames dividing a search display increased the intercept of the visual search function, although in the efficient search task, the divider frames initially did not exert this kind of influence. This result indicates that for a viewer’s first glance at a display, the presence of divider frames can interfere with visual search. Recent studies (e.g., Castelhano & Heaven, 2011; Castelhano & Henderson, 2007; Torralba et al., 2006; Võ & Henderson, 2010) have suggested that global scene information (e.g., gist) is processed quite rapidly and also that it continues to influence subsequent search performance. As we described above, observers immediately confront a visual environment that seems redundant due to the presence of dividers. Therefore, dividing frames initially interfere with visual search.
However, divider frames reduced search slopes in the search for an O. This indicates that divider frames boost search efficiency relative to a single-frame condition in an inefficient search. This facilitation effect does not reflect specific target-oriented guidance, but rather suggests the operation of a guidance mechanism that controls the serial allocation of attention to each framed area. This is because this facilitation occurred not only in the target-present, but also in the target-absent, condition. One reason why the segregation of areas in the four-frame condition elicited both interference and facilitation effects on visual search appears to be that multiple frames create a sorting effect on the items to form groups, and that this affects search performance (cf. Egeth et al., 1984; Kim & Cave, 1999; Niemelä & Saarinen, 2000; Treisman, 1982). This interpretation is consistent with a recent study by Xu (2010), who reported that, relative to nonclustered items, clustered search items (i.e., in groups) facilitated inefficient search (e.g., a T among Ls), whereas they interfered with efficient search (a T among Os). More importantly, the facilitation of visual search by multiple divider frames appears to happen even when the to-be-searched items are uniformly distributed—that is, not clustered. This issue will be considered in more detail in the General discussion. To summarize, at least for inefficient visual search tasks, divider frames within a search display appear initially to interfere with, and then to facilitate, visual search performance.
In Experiment 1, divider frames within a search display were shown to affect visual search, but this depended in part on the type of search task (inefficient or efficient). At least for inefficient searches, the effect of frame dividers appears to resemble the effects of item clustering (cf. Xu, 2010). Furthermore, with inefficient search, Xu reported that as cluster size (i.e., the number of items per cluster) increased, visual search became faster. In Experiment 2, thus, we pursued these effects by manipulating the number of divider frames presented in a search array. In this experiment, we conducted a visual search task in which the overall set size was fixed at 64, and we manipulated the number of frames (one, four, 16, or 64). We then examined the effect of the number of frames on inefficient search.
A group of 20 undergraduate and graduate students (mean age = 22.6 years) participated in Experiment 2. All of them had normal or corrected-to-normal vision and were naive with respect to the purpose of the research. The data of one participant were removed from the analysis because of performance much lower than the overall mean (i.e., mean RTs in the one-frame conditions of about 3.5 s in the target-present trials and about 10 s in the target-absent trials).
In this experiment, we used one set size condition, with 64 items. The items were located on an imaginary 8 × 8 grid (the size of each cell was 2.0° × 2.0°) with a ±0.1° randomly oriented jitter. Thus, the area surrounded by a frame contained 64 items, 16 items, four items, and one item in the one-frame, four-frame, 16-frame, and 64-frame conditions, respectively. In each frame condition, target presence was also manipulated. In the target-present condition, one target O was presented with 63 Cs, and in the target absent condition, 64 Cs were presented.
Apparatus and procedure
The apparatus and task procedures were identical to those of Experiment 1. The participants were instructed to search for an O in the visual search display. They completed an experimental session of 512 trials, with 64 trials (the locations of the target were counterbalanced) in each of the eight conditions created by the 2 (target present or absent) × 4 (one, four, 16, or 64 frames) factorial design. The order of the trials was randomly determined. Prior to the experiment, the participants completed a practice session of 32 trials, with four in each of the eight conditions.
Results and discussion
We conducted an ANOVA with the factors Target Presence and Frame on the RTs. RTs in the target-absent conditions were greater than those in the target-present conditions, F(1, 18) = 77.88, p < .001, ηp2 = .81. Furthermore, the main effect of frame was significant, F(3, 54) = 4.91, p < .01, ηp2 = .21: RTs in the four-frame conditions were faster than those in the other three divider frame conditions, ps < .02. Pairwise differences among the other three frame conditions were not significant, ps > .3. The interaction between frame and target presence was not significant, F(3, 54) = 1.90, p > .1
Performance was better (i.e., faster) in the four-frame than in the one-frame condition (this baseline condition corresponded to a typical visual search condition). This result replicates the findings of Experiment 1. As such, it supports the view that frames that divide a visual search area can facilitate inefficient search. Furthermore, the finding of a null interaction of target presence with frame strengthens the interpretation that divider frames can facilitate the serial allocation of attention to each framed area. However, the performance in the other two experimental frame conditions (i.e., the 16-frame and 64-frame conditions) did not differ from that in the one-frame (baseline) condition. It is not entirely surprising that the frames in the 64-frame condition failed to improve visual search, as these divider frames did not collect items into larger groups. In contrast, it is a little strange that no facilitation was observed in the 16-frame condition, in which groups of four items were arguably formed. These results indicate that there can be an appropriate number of frames to facilitate inefficient search. We discuss this issue further in the General discussion.
In an efficient search, divider frames may generally add unnecessary information to the visual search display. If this is so, in Experiment 3 we would expect to find that increasing the number of frames would interfere with visual search and lead to slower RTs in conditions with more divider frames.
A group of 15 undergraduate students (mean age = 21.1 years) participated in Experiment 3. All of them had normal or corrected-to-normal vision and were naive with respect to the purpose of the research. It should be noted that 14 of the participants had completed Experiment 2.
Stimuli, apparatus, and procedure
The stimuli, apparatus, and task procedure were identical to those of Experiment 2, with the following exceptions: In the target-present condition, one target C was presented with 63 Os, and in the target-absent condition, 64 Os were presented (see Fig. 3). The participants were instructed to search for a C.
Results and discussion
Overall, RTs were slower in the target-absent than in the target-present conditions, F(1, 14) = 30.86, p < .001, ηp2 = .69. Furthermore, the main effect of frames was significant, F(3, 42) = 4.13, p < .02, ηp2 = .09. RTs were slower in the 16-frame than in the one-frame conditions, p < .01, but pairwise differences among the other conditions were not significant, ps > .1. The interaction between target presence and frame was not significant, F < 1.
Although the performance in the four-frame condition was not significantly different from that in the one-frame condition, performance was slower in the 16-frame than in the one-frame condition. This result indicates that the frames dividing the visual search area did interfere with efficient search. If this interference effect simply resulted from the cost of ignoring the unnecessary information in the visual search task, its magnitude should have increased with increments in the unnecessary information (i.e., the number of frames). In this context, the fact that our results revealed that the 64-frame did not interfere with efficient search is interesting, and perhaps surprising. In simple terms, the prediction that the interference of divider frames should be greater in the 64-frame than in the 16-frame condition may be reasonable, if the major causal factor is an increment in unnecessary information. However, the results are inconsistent with this prediction. Instead, they show that although unnecessary frame information that outlines item groups (as, e.g., in the 16-frame condition) can interfere with efficient search, more unnecessary frame information that does not collect items into a group-like form has no negative impact on visual search. This finding is consistent with the suggestion that when faced with grouped stimuli, our visual system is drawn automatically to global aspects of a stimulus array, and that this, in turn, can interfere with direct access to the individual items that comprise the grouped configuration (e.g., Kimchi, 1992; Navon, 1977).
In the present study, we examined the effects of frames that divide a visual search area into smaller areas on both inefficient and efficient visual search. Taken together, our results suggest that divider frames initially interfere with visual search, due to information redundancy (Exps. 1 and 3), but that they subsequently facilitate visual search by guiding the serial allocation of attention to each framed area (Exps. 1 and 2). These results are consistent with previous studies that have shown that visual search can be influenced by both scene gist (e.g., Castelhano & Heaven, 2011; Castelhano & Henderson, 2007; Torralba et al., 2006; Võ & Henderson, 2010) and perceptual grouping (e.g., Egeth et al., 1984; Kim & Cave, 1999; Niemelä & Saarinen, 2000; Treisman, 1982; Xu, 2010). Visual search behaviors can be influenced by grouping on the basis of spatial proximities (Xu, 2010) or of similarities among the stimulus features (e.g., Egeth et al., 1984; Kim & Cave, 1999; Niemelä & Saarinen, 2000; Treisman, 1982). In addition, we reported here that a completely enclosed subarea containing multiple items, by itself, can influence visual search by creating a grouping effect.
Nevertheless, this effect of the frames on search performance may differ from the conventional effects of perceptual grouping that have been reported in previous studies. This is because the divider frames do not fundamentally change the visible appearance of a search display. For example, at a glance, we can discriminate displays that contain clustered items from displays containing uniformly distributed items (Xu, 2010); the spatial layouts of items in these two displays are dramatically different.2 In contrast, the frames used in this study simply added unnecessary information to the background of the visual search display; they did not alter the spatial layout of the search items. Yet this type of imposed grouping did affect visual search performance.
This study shows that spatial areas separated by dividers initially exert an interfering effect on visual search, and subsequently serially guide attention. This is consistent with a recent suggestion by Wolfe, Võ, Evans, and Greene (2011) regarding the nature of visual search in scenes; they proposed a dual-path model for visual search in scenes. According to this model, visual search may be accomplished by a “selective” path, in which objects are individually processed, and by a “nonselective” path, in which global information is extracted. In this view, the dividers segregating a visual display are first processed in the “nonselective” path as hindrances. Thus, initially these divider frames would interfere with visual search performance when search is a relatively easy task (i.e., efficient search). However, each of the dividing frames would later be processed individually, in the “selective” path, as an object collecting some items into one group, and could be regarded as a guide for inefficient search.
Although it is beyond the scope of this study to compare the absolute impacts of each perceptual grouping, here we want to discuss and speculate about this issue. It is conceivable that the effect of the frames on visual search is relatively small. One reason for this belief is that, if the perceptual grouping determined attentional width in a stimulus-driven way, the sizes of the attentional foci that would have to grasp a set of grouped items in a single attentional fixation would differ in these two scenarios. The zoom lens model, which is one dominant theory of attention, suggests that processing efficiency for stimuli within a focused area increases as the area of the attentional focus decreases, and vice versa (e.g., Castiello & Umiltà, 1990; Eriksen & St. James, 1986; Maringelli & Umiltà, 1998). This suggestion could be applied to visual search (e.g., Bravo & Nakayama, 1992; Greenwood & Parasuraman, 1999, 2004). For example, in Xu’s (2010) experiments, more items were concentrated in a small area in a clustered than in a nonclustered condition. Thus, according to the zoom lens model, observers could attend to more items at one fixation in the clustered condition using a small-sized attentional focus. Therefore, this type of spatial grouping might improve inefficient search performance not only by facilitating the serial allocation of an attentional focus, but also by increasing the processing efficiency within the focus itself (i.e., more items could be processed in a single fixation). In contrast, the spatial layout of the component items in a visual search area was not altered by the dividing frames in this study. In this situation, observers cannot attend to more items with a small-sized attentional focus, because the spatial proximities of the items remain unchanged. Therefore, the improvement of visual search performance in this study may largely reflect facilitation of the serial allocation of attention. For this reason, the magnitude of the effect of the divider frames on visual search may be relatively small.
On the basis of this discussion and speculation, we further speculate that an optimal grouping effect may be conferred by divider frames for different types of visual search tasks. Several studies have proposed that some items, rather than one item, can be processed simultaneously in a single attentional fixation (e.g., Humphreys, Quinlan, & Riddoch, 1989; Pashler, 1987). Taking into account a suggestion from the zoom lens model of attention (e.g., Eriksen & St. James, 1986), the number of items processed at one time may depend on the complexity of the stimulus items. That is, observers can process many simple items at one time, but they cannot process many complex items. Moreover, it is well known that the time scales of visual search tasks differ as function of the task type, from a signal detection task in a laboratory to a visual search task in a natural scene (e.g., Wolfe, 1998a, 1998b, 2010). These differences may result from differences in processing difficulties. That is, more complex stimuli (e.g., when the target and distractors are difficult to discriminate; Duncan & Humphreys, 1989) require more attentional resources to process (i.e., more time). Considering the relationship between the size of the attentional focus and processing efficiency, it is possible that an ideal attentional width (or the number of items focused on in a single attentional fixation) exists for particular stimulus types. This optimal grouping may occur when the size of a frame matches the ideal focal attentional width.
To summarize, because dividers separate a visual area into smaller subareas, which is typical in our everyday visual environment, their effects should be assessed in order to fully understand how vision operates in the world around us. Considering that we have many chances to search for a target object in a complex environment (i.e., an inefficient search task), our study suggests that finding such a target object within an array of other objects that is segregated by dividers would be more efficient in our daily lives. However, many issues related to visual searches in divided areas have yet to be clarified. For example, among other things, it is unclear whether the dividers influence other types of search tasks (e.g., search for a T among Ls, search for a “natural object” in a natural scene task, and so on). Furthermore, the sizes and shapes of frames as well as their relative homogeneity may matter in a given search display. In order for us to understand our routine visual searches better, such issues are interesting and important, not only for experimental but also for applied psychology. We intend to examine these issues in more detail in future research.
The ANOVA did not reveal a difference between the one- and four-frame conditions in the search-for-C task. However, when we divided the data into two groups (the target-present and target-absent trials), separate t tests showed that the four-frame condition increased the intercept in the target-present trials, t(11) = 2.79, p < .02, whereas it did not do so in the target-absent trials, t < 1. Therefore, we do not strongly suggest that the dividers (i.e., four frames) never interfere with efficient search. We examined this issue further in Experiment 3.
Other types of perceptual grouping (e.g., by similarity) do, of course, change the appearance of a visual display, as compared to a display in which items are distributed randomly. Thus, here it is obvious that the type of perceptual grouping can influence visual search.
R.N. is now at Tohoku University as a postdoctoral researcher. This study was supported by a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science, awarded to K.Y.