Abstract
Visual search in traditional, abstract arrays is typically dominated by feature guidance. In contrast, search in natural scenes is thought to be dominated by scene-level guidance, both semantic (knowledge of the plausible regions where an object might be found) and episodic (knowledge of where a specific exemplar is located). Wolfe et. al (2011) proposed a two-pathway architecture for search in scenes. The dominant pathway is 'nonselective' and utilizes scene-level guidance, derived from gist recognition, to direct attention to plausible regions. The secondary 'selective' pathway uses feature guidance to further narrow search. However, this claim of a hierarchal relationship (initial gist-based guidance, followed by local feature guidance) has not been tested directly. The present study investigated this relationship, and we hypothesized that feature guidance plays a more substantial and earlier role than currently theorized. We placed scene-level guidance in conflict with feature guidance from a search-task-irrelevant color maintained in visual working memory (VWM). Participants searched for a letter superimposed on an object in a natural scene, receiving no cue, a word cue, or a picture cue for the target object. Simultaneously, they maintained a color in VWM for a later memory test. Half of the scenes contained a distractor object with a color that either matched or mismatched the remembered color. Search times declined with increasing cue specificity. Importantly, under all conditions of guidance, participants were more likely to fixate the distracting object on match compared with mismatch trials, increasing search times. This feature-based capture was most pronounced during the first few saccades on the scene. Moreover, capture occurred even on trials where the distracting object appeared in an implausible location for the target. These results imply a substantial role for feature guidance in the earliest stages of real-world search. Search through natural scenes is not necessarily dominated by gist-based guidance.
Meeting abstract presented at VSS 2016