Zusammenfassung
Our V1 model (Li, PNAS 1999, Spatial Vision 2000) is based on physiological evidence of contextual influences instantiated by horizontal connections between nearby neurons tuned to similar orientations. The contextual influences are thought to underlie the segmentation of visual scenes by enabling relatively higher responses to visual features at or near the boundaries between image regions. The model is able to segregate two regions of texture that differ only in the mean orientation of the elements within them (figure, left). The perceptual location of the boundary is determined by the locations of the texture bars which induce higher responses. Due to facilitation between co-aligned bars and iso-orientation suppression within a homogeneous texture, V1 responses are relatively higher to texture bars co-aligned with, and close to, the texture boundary (see figure, where higher model responses are indicated by thicker bars in model outputs). The model therefore predicts that a texture boundary will appear closer to the texture region whose elements are co-aligned with the boundary. Here we show that this predicted perceptual bias, up to 0.8 units of texture element separation, agrees well with human vision (figure, right). We explore the effects of orientation, contrast and density of the texture elements on this bias.
