Size and Position Invariance in the Visual System

Abstract

The encoding of simple features such as lines and angles leads naturally to size and position invariance when these features are analyzed symbolically in terms of structure and connectedness. No such analysis is possible for a spatial frequency encoding, however, as all a priori indications of pattern unity are lost. An additional transformation of spatial frequency information is therefore demonstrated that automatically derives a size invariant representation of form. The feasibility of the transform scheme in the visual system is outlined with a hypothetical processing sequence. It is found that the organization of cells in the striate cortex shows a marked similarity to an intermediate stage of the proposed sequence.