Optical Pattern Recognition Using Synthetic Discriminant Functions

摘要

This thesis analyzes the applicability of the matched spatial filter to optical pattern recognition when the object has been rotated relative to the filter. Several techniques are introduced to enhance the discrimination efficiency of the matched filter. Energy normalization was applied to compensate for the variance in the reflected energy of targets. This allowed the matched filter to detect the desired target in the presence of bright background noise. A rotation invariant matched filter was constructed using multiple exposure holograms. Correlation averaging was also applied to make the filter rotation invariant.