Spatial competition on the master-saliency map

Abstract

The saliency map model (Itti and Koch, 2000) is a hierarchically structured computational model, simulating visual saliency processing. Iso-feature processing on feature maps and conspicuity maps precedes cross-dimensional signal processing on the master map, where the most salient location of the visual field is selected. This texture segmentation study focuses on a possible spatial structure on the master map. In four experiments the spatial distance between a texture irregularity in the stimulus (“target”) and a cross-dimensional task irrelevant texture irregularity in the backward mask (“patch”) was varied. The results show that the target-patch distance modulates target detection, and that this modulation is limited to critical distances around the target. We conclude that the signals from different feature dimensions compete on a spatial master map. There is first evidence that the critical distances increase with target eccentricity.