Some fingerprints of V1 mechanisms in the bottom up saliency for visual selection

Li Zhaoping, Keith A. May, Ansgar Koene Department of Computer Science, University College London, UK To appear as Chapter 7 in Computational modelling in behavioural neuroscience: closing the gap between neurophysiology and behaviour Editted by Dietmar Heinke and Eirini Mavritsaki, Psychology Press, 2009. Abstract: A unique vertical bar among horizontal bars is salient and pops out perceptually regardless of the observer’s goals. Physiological data have suggested that mechanisms in the primary visual cortex (V1) contribute to the high saliency of such a unique basic feature, but fail to indicate whether V1 plays an essential or peripheral role in inputdriven or bottom up saliency. Meanwhile a biologically based V1 model has suggested that V1 mechanisms can also explain bottom up saliencies beyond the pop out of basic features (Li 1999a, 2002). For instance the low saliency of a unique conjunction feature like a red-vertical bar among red-horizontal and green-vertical bars is explained, under the hypothesis that the bottom up saliency at any location is signalled by the activity of the most active cell responding to it regardless of the cell’s preferred features such as color and orientation. While some recent experimental data have provided support for this V1 saliency hypothesis, higher visual areas such as V2 and V4 also contain neurons tuned to similar basic features that can pop out in the bottom up manner. Furthermore, previous saliency models can capture much of the visual selection behavior using generic rather than V1 specific neural mechanisms. It is therefore important to ascertain V1’s role in saliency by identifying visual selection behavior that show specific identifying characteristics, i.e., fingerprints, of V1 or other cortical areas. In this paper, we present our recent findings on bottom-up saliency based behavior of visual search and segmentation that directly implicate V1 mechanisms. The three specific fingerprints are: (1) ocular singleton captures attention despite being elusive to awareness, (2) V1’s collinear facilitation manifested in texture segmentation, and (3) a match between the redundancy gains in double feature singleton search and V1’s conjunctive cells.