Tisha Reid, M.A. and Lauren Scharff, Ph.D.
Stephen F. Austin State University
Background and Purpose
Visual search occurs while one is looking for specific items in a complex visual scene. This type of task is commonly performed in the real world. Researcher's endeavors to better understand visual search have revealed two types of processing that typically occur for such tasks (e.g. Beck, 1982; Bergen and Julesz, 1983; Julesz, 1984; Kaplan, Carvellas, 1965; Kaplan, Carvellas, Metlay, 1966; Selfridge, 1959; Treisman, 1986; Treisman and Gelade, 1980; Treisman and Gormican, 1988). In parallel searches, targets are located effortlessly; thus, the targets appear to "pop-out" among the distracters regardless of the number of distracters present in the display. Serial searches require the participant to attend to each individual object in the display until the target is located. Reaction times for serial searches are dependent upon the number of distracters.
Generally, visual search tasks studied in the laboratory have utilized artificial, unnatural stimuli which include very simplistic targets on blank backgrounds. Using stimuli that were more complex and continuous, and thus more naturalistic, Wolfe (1994) found that conjunctive search on complex backgrounds can be highly efficient. He concluded that models of visual search based on laboratory data can be generalized to “real world” situations. An additional naturalistic feature, three-dimensionality, was investigated by Enns and Rensink (1990). They found that rapid, preattentive processing is not limited to simple, non-conjunctive features which imply 2-D perspective.
The current study employed more naturalistic, continuous stimuli, similar to Wolfe (1994). However, unlike Wolfe, the current stimuli also included conditions with 3-D cues such as shading, linear perspective, and direction of lighting. This experiment used a forced-choice method, mixed design to investigate the effect of target type (2-D vs. 3-D), background type (blank, complex, or complex with cues), number of distracters (5, 10, or 15), target presence or absence, and implied lighting direction (left or right) on visual search. Figure 1 illustrates the six basic target type and background condition combinations.
Trials were blocked by direction of lighting and background. Thus, there were 6 separate blocks for each target type (between subject variable), the order of which was randomized across participants. The number of distracters was randomized within a block. The major dependent variable of interest was the reaction time (the time required by the participant to respond that the target was present or absent) but number of errors was also recorded. The target-present and target-absent experimental trials each consisted of 10 different presentations of the 3 possible display densities, leading to 60 trials per block. Thus, participants (N=32) ran a total of 360 experimental trials (30 with targets present and 30 with targets absent for each of the 6 blocks). In addition, participants received 12 practice trials prior to each block of conditions.
Participants indicated target presence or absence by pressing one of two keys on the keyboard. Both speed and accuracy of responses were stressed to the participants.
Figure 1: Example stimuli: 2-D bars and 3-D cubes on blank, complex, and complex-with-cues backgrounds. (Note: picture fuzziness is due to jpeg image manipulation; the experimental stimuli were not blurred.)
Contrary to prediction, the main effect for target type was not significant (F(1,30)= .038; p=.846). Significant main effects were found for number of distracters (F(2,60)=199.5; p=.000) and between present and absent trials (F(1,30)=241.5; p=.000). Reaction time increased as number of distracters increased, and target-absent trials took longer than target-present trials. Further, a significant interaction occurred between present/absent and number of distracters (F(2,60)=137.8; p=.000); target-absent trials showed a steeper rise in reaction time as compared with target-present trials. Figure 2 shows reaction time as a function of target type, number of distracters and present/absent.
A significant main effect was found for background (F(2,60)=241.5; p=.0008); the complex background had the fastest reaction time followed by the complex background with 3-D cues, and finally the blank background. Importantly, background did not interact with the variables present/absent and number of distracters; see Figure 3. Thus, findings with black backgrounds may be generalized to more complex backgrounds.
However, the simple main effect of background was significantly modified by target type (F(2,60)=7.3; p=.0014). The reaction time for 3-D targets did not vary across background conditions, while much variability existed for the 2-D targets. The search for 2-D targets was slower than the search for 3-D targets on a blank background, while it was much faster on a complex background, and somewhat faster on a complex background with 3-D cues.
Finally, as expected, there was no main effect for direction of lighting (F(1,30)=.14; p=.715); however, it did show high level interactions by modifying the reaction time to target-absent trials in conditions with 15 distracters.
The above main experiment used several subjects, but did not allow for the large data collection set traditionally used for visual search tasks. Perhaps the absence of a significant main effect of target type was due to the lack of participant practice with the task. If a search easily leads to pop-out, then practice should not significantly influence reaction time compared to non-pop-out tasks. However, recent empirical evidence suggests that individual differences may influence reaction time for visual search (e.g. Treisman, 1986; Treisman and Gelade, 1980). Thus, processing styles for certain stimuli may not be strictly parallel or serial.
In order to test this secondary hypothesis, three participants completed 5 complete data sets. The first set was solely for practice with the task. Average reaction times and standard errors for the final four data sets are illustrated in Figures 4-6 for these three subjects.
When these figures are compared to the group data shown above, the following major points are apparent. First, even with practice, there are individual differences in the processing across target type. While all three subjects show serial processing of the 2-D bars, some subjects show a more parallel processing trend for the cubes (strongly parallel for LVS, somewhat parallel for SRH, and serial for TMR). Second, regardless of processing across target type, all three subjects showed no effect of background. Thus, as with the group data, the implication is that results using black backgrounds may be generalized to more complex backgrounds.
The main experiment suggested serial processing for 2-D bars and 3-D cubes, both of which contain conjunctions of features. However, the results of the control experiment indicate that for some individuals the implied depth cues can lead to parallel processing. This latter finding supports Enns and Rensink's (1990) conclusion that scene-based properties such as implied 3-dimensionality may allow preattentive processing. Other researchers (e.g. Wolfe, Cave, and Franzel, 1986; Nakayama and Silverman, 1986) have also found evidence for parallel processing of conjunctions of features. Further, the target type results suggest that individual differences occur, so that certain tasks may not strictly lead to serial or parallel processing.
Finally, the processing trends across background do not significantly differ. Thus, as also concluded by Wolfe (1994), results from blank background conditions seem to generalize to more complex background conditions.
We plan to investigate the effects of search area size, the use of color versus gray backgrounds, and top v. bottom implied lighting compared to left v. right implied lighting as possible influences in the processing of target type.
Back to Research Interests.