Cortical anatomy, size invariance, and spatial frequency analysis.

Abstract

In a recent application of an algorithm developed in computer and optical pattern recognition, Cavanagh has suggested that a composite of spatial frequency mapping and complex logarithmic mapping would provide translationally, rotationally, and size-invariant mechanism for human vision. In this work, Cavanagh has not made explicit the fact that this transformation is composite, that is, that the first step (global Fourier analysis) perceptually, anatomically, and physiologically inconsistent with primate vision, but that the second step (complex logarithmic mapping) is actually embodied in the anatomy of the primate retinostriate projection. Moreover, it is the complex logarithmic remapping step which is entirely responsible for the computational simplification of the symmetries of size and rotation invariance. These facts, which have been extensively discussed in a recent series of papers, are briefly reviewed and illustrated. Furthermore, it is shown that the architecture of the retinostriate map may provide an example of computational anatomy in vision, such that the spatial representation of a stimulus in the brain may be of direct functional significance to perception, and to the nature of certain visual illusions.