A theoretical and computational investigation into aspects of human visual perception : proximity and transformations in pattern detection and discrimination

Abstract

A variety of measures are enlisted in an explanation of some longstanding perceptual phenomena associated with an assortment of visual patterns. In following the proximity principle of Gestalt psychology, these are commonly based upon a statistical treatment applied to one or another of a hierarchy of distance measures. Following from this, some problems of visual perception are tackled in terms of an active perceiving mechanism, which generates transformations in the realization of object invariance in space and over time. This generative transformational approach is also employed in an account of perception of various patterns and visual illusions. Although a range of proximity measures is involved throughout, the nearest neighbour metric is staple. For perception of unstructured visual arrays, the contribution of distance mechanisms, particularly nearest neighbours, is shown to be important. For structured arrays, the contribution of distance mechanisms along with transformations is important. Information about relative positions of image elements permits the selection of transformations that reveal structure. With respect to such information, however, the proximity principle is taken to its limits.