Abstract

Composite filters, which are linear combinations of matched filters, allow arbitrary correlation peaks to be obtained for any objects but are usually associated with large sidelobes. This problem Is compounded when the composite filters are composed of circular harmonic rotation-invariant filters,¹ because such filters trade off the SNR for rotation invariance. One way to reduce sidelobes is to make a filter that minimizes the mean-square difference between the outputs of two filters: One Is designed to recognize a given object and to discriminate against another given object. The other Is designed to discriminate against noise. The combined filter yields good sidelobe reduction.

© 1987 Optical Society of America