A Pattern Recognition Program which uses a Geometry-Preserving Representation of Features

Abstract

This report describes a two layered pattern recognition program. The first layer scans an input for features and produces a coded representation. The second layer looks for combinations of code which signify relations between features in the input. It is argued that the nature of the representation produced by the first layer determined the amount and quality of subsequent processing. Both layers create and modify their operators. Good results are obtained using the Highleyman hand printed data. The program is discussed as a model of an aspect of human perception.