A Theory of Pattern Perception Based on Human Physiology

Abstract

The gross connectivity, patterns of information pathways in the primate and human visual systems, when examined by an information processing engineer, bear a curious resemblance to the two-dimensional-pattern optical computers which he builds himself in an attempt to achieve pattern recognition. The portions of the visual system inferior to the primary visual cortex are essentially a topologically accurate homeo-morphic mapping of the retinal image, at least in the vicinity of the fovea. Analysis of the topological aspects of the visual scene and hence ‘ Pattern Recognition ’ must therefore take place in the primary visual cortex and successive cortical areas We have previously reported how intra- and inter-cortical connectivity could support a combined memory and computation scheme capable of performing pattern recognition by a variation of two-dimensional cross correlation. This paper is a report of an extension of the previous model enabling it to perform pattern recognition by computing the two-dimensional Fourier transform of input images in a manner isomorphic to computation of the Fraunhofer diffraction pattern in optical computers. It is shown that the use of the Fourier transform of an unknown pattern in a subsequent correlation scheme results in a pattern recognition system which is not easily faulted by the small local mutilations of input patterns which badly compromise straight correlation pattern recognition schemes