A suppression template for multiple distractors in visual search

To search efficiently, it's ideal to pursue a goal-relevant object (i.e., target) without any distraction from goal-irrelevant objects (i.e., distractors). In reality, however, it's almost always necessary to filter out distractors. Previous research showed people create a distractor template for rejecting specific distractors (Arita et al., 2012). However, it's still unclear if the distractor template can be made for multiple distractors simultaneously. Doing so would be advantageous for complex environments in which distractors outnumber targets. We investigated this question using a visual search task in which participants searched for a gray (target) square among colored distractors (see supplemental figures). During "training", participants saw the same three colored distractors either simultaneously (Experiment 1,2,4) or sequentially (Experiment 3). During "testing", two new distractors sets were interleaved with the trained distractors. The critical manipulation in each study was based on the distance (in color space) of the new "test" distractors from the learned "trained" distractors. We hypothesized that if the trained distractor template created is inclusive enough to suppress the new "test" distractors, then search performance should be just as good with the new distractors as the old ones. In contrast, if the template is specific to the trained colors, then performance should be worse. Experiment 1 found no performance cost for new distractors near the outer range of trained distractors, but significant cost for distractors farther away. Experiment 2-3 found that new distractors within the trained color range also showed no performance cost, and replicated the finding that distractors farther outside of trained color range interfered with performance. Finally, in Experiment 4 we found that the distractor template was not narrowly defined by the mean of three trained colors. This study suggests that multiple distractors create a broad template that suppress all feature values within (and very near) the trained color range. Meeting abstract presented at VSS 2017