Understanding conjunction and double feature searches by a saliency map in primary visual cortex.

By Li Zhaoping , presented as Vision Science Society Annual meeting, Sarasota, Florida, May 10-15, 2002

Visual search is the task of finding a target among distractors. When the target has a feature that is absent in the distractors, the search can be very efficient and termed feature search when the feature is in a basic feature dimension like color, orientation, depth, and motion direction (Treisman and Gelade, Cog. Psychol. 1980). When the target is only distinguishable by a particular conjunction of features, e.g., green and vertical, each of which is present in the distractors, the search is termed a conjunction search. Some conjunction searches, e.g., conjunctions of depth-orientation (Nakayama and Silverman, Nature 1986) and motion-orientation (Mcleod et al, Nature 1988), can be efficient, while others, such as color-orientation, may be very difficult depending on the stimuli (Treismand and Gelade 1980, Wolfe, Vis. Res. 1992). Double feature searches are those for which the target differs from distractors in more than one feature dimensions, e.g. a green-vertical target bar among red-horizontal distractor bars. They should be no less efficient than the two corresponding single feature searches (e.g., green target among red distractors or vertical target bar among horizontal distractor bars). The double feature advantage is stronger for some double features, such as motion-orientation, than others, such as color-orientation (Nothdurft Vis. Res. 2000). I use a V1 model to show how various efficiencies in these search tasks can be understood from a saliency map in V1 (Li, TICS 2002). Contextual influences make V1 response increase with stimulus saliency which determines search efficiency. The degree of availability of V1 cells tuned to various conjunctions of features, and the tunings or specificities of the intra-cortical connections to the optimal feature values of the pre- and post-synaptic cells, are shown in the model to dictate the search efficiencies.