Testing a V1 model --- Perceptual biases and saliency effects

Abstract

Background: a preattentive segmentation model using contextual influences in V1 predicts certain perceptual biases in the location of a texture border and variations in the border saliency (Li, Spat Vis 2000). Specifically, when two textures composed of iso-oriented elements border each other with a given orientation contrast at the boundary, the perceptual location of the border is biased towards the texture whose elements are aligned with the border. The saliency of the border increases with the increased alignment of the elements from either texture region with the border. Purpose: to validate the model by measuring these biases and saliency variations. Method: thresholds and biases for the horizontal alignment of vertical texture boundaries were measured in 4 observers using the method of constant stimuli. Observers judged, on each trial, whether the texture boundary in the upper of two vertically separated panels was left or right of the texture boundary in the lower panel. The angle between the Gabor patch elements and the boundary was varied from 0–90 deg, while the orientation difference at the boundary was a constant 90 deg. Results: bias toward the aligned boundary elements was, at 0.3 deg, almost as large as the inter-element separation. Thresholds were highest at 30–60 deg. Conclusions: the results confirm the preattentive segmentation model. Generalization to different scales and orientation contrasts can help establish the parameters of the model for human V1. Psychophysical tests of other related predictions from the V1 model will also be presented.