Abstract

There are different requirements on the specificity of a correlation filter used for optical pattern recognition that depend on the particular application. For example, in industrial inspection, small differences between objects are important so the filter should be designed to be very specific. In contrast, target recognition often requires detection of a class of similar object. We have previously designed correlation filters, using a spectral iteration technique, that are invariant to perspective view, position, and intensity while remaining specific to the target object. In operation, several such holographic filters are applied to the input scene, and as the filters are changed, the output is processed for points of constant intensity. These points identify target locations. We now apply different constraints during the iteration process to modify the specificity of the filters. Methods include using fewer harmonics, apodizing the eigenvalue profile, and forcing certain harmonics to zero. We apply the filters to blurred, noisy, or partially obstructed objects.

© 1987 Optical Society of America