Effect of disjoint illumination in optimal correlation filters with obscured inputs

Abstract

We evaluate the target recognition performance of five optimal filter correlators for targets that are partially obscured by a disjoint background pattern of constant illumination. The optimal filters are based on cascading a matched filter with a correction filter to optimize a selected metric. In the two correlators with superior performance, the correction filter includes a term that correlates a phase extracted portion of the template with a phase extracted portion of the input signal. We also propose a restriction on these optimized correction filters, which preserves the conventional sense of template matching.