Abstract

A learning process is essential in constructing synthetic reference patterns for pattern recognition. A knowledge base is generally built up or updated through correlating adaptively changed reference patterns with a given standard. The optical correlator provides an alternative in the learning process. Compared to learning by a pure digital process, an optical approach substantially reduces the necessary learning time through high speed Fourier transforms and parallel processing. In this paper, we present an adaptive learning process for binary patterns using optical correlators. A direct search algorithm is employed in the learning process to ensure that an optimum pattern can be reached. The variation of correlation peaks obtained from a given input object and adaptively changed reference patterns are used to determine an optimum synthetic reference pattern. The approach is verified by the results obtained from a full range of computer simulations. The ability of the presented algorithm for tolerating noise is also demonstrated.

© 1992 Optical Society of America