Does visual search involve a salience map?

Abstract

Most visual search theories (e.g., Itti & Koch, 2000; Treisman & Sato, 1990; Wolfe, 1994) suggest that signals from each perceptual dimension (e.g., color, orientation, size) are first analyzed separately and then integrated onto a “salience map”. The notion of a salience map is central to accounts for many search phenomena. A core prediction of the use of a salience map is that across-dimension noise should impair a within-dimension search. However, in most situations across-dimension distractor heterogeneity does not reduce search efficiency (e.g., Treisman, 1988). In this study, we examined factors that may underlie this absence of an effect. In four experiments participants engaged in visual search tasks with target dimensions including color, orientation and size. We found that across-dimension heterogeneity did not influence search efficiency when we varied task (detection or compound search tasks), target salience, or task relevance of the non-target dimension. These results suggest that the lack of an acrossdimension effect was not due to a ceiling effect (i.e., noise being too weak to impair a pop-out search) or top-down suppression to the non-target dimension. It appears that integration of dimensional signals does not occur naturally. However, in a situation where across-dimension distractor heterogeneity was introduced alongside heterogeneous within-dimension distractors, search efficiency was reduced. This suggests that the salience map may be engaged only when it is deemed necessary; for example, when dimension-specific signals do not guide search effectively. Taken together, these results suggest that visual search may be primarily driven by dimension-specific maps. Across-dimension signals may be integrated only when dimension-specific guidance is absent.