Abstract

This paper reviews recent research in rotationinvariant pattern recognition and classification methods using mostly circular harmonic matched filters. The early methods using single CHC filters are inadequate for classification of similar objects such as aircraft on cluttered backgrounds. The performance of such filters may be dramatically improved by combining several filters together in a single composite filter or by combining the results of several filters used sequentially. Methods of tuning the filters to eliminate sidelobes and clutter are presented, and implementations for real-time optical processing systems are discussed. Among the possibilities for the latter are the use of a new kind of computer-generated computer hologram for spatial light modulators with a small number of degrees of freedom, of binary phase-only CHC filters for rapid spatial light modulators, and of joint-transform Fourier-plane thresholded optical correlators. Because all the methods are based on matched filtering, they may be implemented either by purely digital means or by means of a hybrid system using an optical correlator with spatial light modulators and digital analysis of the output.

© 1988 Optical Society of America