Seeing circles: what limits shape perception?

Abstract

Humans are remarkably adept at judging shapes and discriminating forms. Forms and shapes are initially sampled by discrete localized visual filters (or receptive fields) in 'early' stages of visual processing. However, more complex higher level filters which integrate or pool information from many local filters may be needed to discern shapes. In order to understand the mechanisms that limit shape perception we asked observers to detect distortions in the shape of briefly presented circles constructed out of samples (Gabor patches, which are well matched to the early visual filters), and varied the radius of the circle, and the number and orientation of the samples. Our results show that shape perception is determined by two factors: the primary determinant is the separation between the samples; however, the orientation of the samples can modulate performance. At small separations, performance is best when the samples are aligned with the global shape, poorer when they are orthogonal, and intermediate when they are all horizontal or vertical. At larger separations these contextual differences disappear; however at all separations, performance is reduced when the orientations of the samples are mixed (i.e. each sample is randomly either aligned or orthogonal, or randomly either horizontal or vertical.). These results suggest an important role for sample separation in shape perception for sampled shapes and suggest that the mechanisms involved in feature binding may modulate the responses of the mechanisms underlying shape perception.