A Theory of Visual Pattern Perception

Abstract

A theory of visual pattern perception is proposed, which is intended to explain first how patterns are learned or committed to memory, second how these patterns are recognized when subsequently encountered, and third how the patterns are recognized under unfavorable real-world conditions, for example, when they are distorted, enlarged, or rotated, or are viewed along with other patterns in a cluttered and noisy visual field. The essential idea of the theory is that each pattern is represented in memory as a network of memory traces recording the features of the pattern and the attention shifts required to pass from feature to feature across the visual field. These attention shifts may take the form of saccadic eye movements or they may be executed internally, according to the angular displacement involved. Memorizing and recognizing a pattern are thus seen to be closely analogous to memorizing and repeating a conventional sequence of behavior, each being an alternating sequence of sensory and motor activities. From this analogy come certain predictions concerning the presence of scanpaths in eye movements during pattern perception, and one of these predictions has been verified experimentally.