Saliency maps for finding changes in visual scenes?

Abstract

Sudden changes in the environment reliably summon attention. This rapid change detection appears to operate in a similar fashion as pop-out in visual search, the phenomenon that very salient stimuli are directly attended, independently of the number of distracting objects. Pop-out is usually explained by the workings of saliency maps, i.e., map-like representations that code for the conspicuity at each location of the visual field. While past research emphasized similarities between pop-out search and change detection, our study highlights differences between the saliency computations in the two tasks: in contrast to pop-out search, saliency computation in change detection (i) operates independently across different stimulus properties (e.g., color and orientation), and (ii) is little influenced by trial history. These deviations from pop-out search are not due to idiosyncrasies of the stimuli or task design, as evidenced by a replication of standard findings in a comparable visual-search design. To explain these results, we outline a model of change detection involving the computation of feature-difference maps, which explains the known similarities and differences with visual search.