Primary visual cortex as a saliency network for pre-attentive segmentation

Stimuli outside classical receptive fields have been shown to significantly modulate neural responses in primary visual cortex (V1). We propose that these contextual influences enable V1 to respond more strongly to important input areas, such as region boundaries, and smooth contours or small figures against backgrounds, as illustrated in figure 1. Such areas therefore become more salient, maybe even `popping out' in a form of pre-attentive segmentation. The cortex detects these important or conspicuous locations by detecting the breakdown of spatial homogeneity in input, for instance, at the border between two texture regions of equal mean luminance. The sudden change in input homogeneity causes changes in the context surrounding any local feature (e.g., a small bar). This results in changes of contextual influences, which often lead to higher responses to these local features. This proposal is implemented in a biologically based model of V1, where contextual influences are mediated by intra-cortical horizontal connections. The behavior of the model is demonstrated using examples of texture segmentation, figure-ground segregation, asymmetry in visual search, and contour enhancement; many of these phenomena have generally been considered products of higher cortical areas. The model predicts that cell responses are tuned to orientations of nearby texture borders relative to their preferred orientations of stimuli within their classical receptive fields. We compare the model behavior with the psychophysical and physiological data, and show how much of the two bodies of data can be linked by this model.

Presented as Poster number 10 (page 104 in the abstract book) at 3rd Annual Vision Research Conference --``Pre-attentive and Attentive Mechanisms in Vision", May 7-8, 1999, a two day satellite symposium prior to the 1999 meeting of ARVO. Fort Lauderdale, Florida, USA. To see the full abstract with figure --- full abstract with figure