Discrete internal pattern representations and visual detection of small changes in pattern shape

Abstract

A scheme is proposed that relates the internal representation of pattern stimuli to visual pattern discrimination performance. The main assumptions of the scheme are: (1) internal representations are assigned to patterns probabilistically; (2) internal representations are discrete; (3) visual discriminability of two patterns is determined by the differences in the assignment probabilities of their internal representations. It is suggested that the internal representations associated with a pattern may be analyzed by measuring the visual detectability of small changes in pattern shape at points along a continuum of shapes. This continuum is generated by a group of transformations smoothly parameterized by a single variable. In two experiments designed to test this approach, subjects discriminated pattern displays containing combinations of three-dot subpatterns. Under the hypothesis that there were just two relevant internal representations, specifying dot collinearity and dot noncollinearity, the qualitative characteristics of the discrimination performance were correctly predicted.