Optimal tradeoff composite correlation filters

In the field of two-dimensional pattern recognition, correlation filters have been used to compare a test image with a known reference image. With the advent of composite filters, multiple versions of an object formed into a training set were used to design a single filter. However, distortions not represented in the training set used in the design of the composite filter would not be recognized as easily. By optimizing individual criteria, early composite filters would attempt to improve different aspects of filter behavior. Recently, the concept of optimal tradeoff filters was introduced that combines multiple criteria into a single optimization. This thesis introduces the Optimal Tradeoff Synthetic Discriminant Function (OTSDF) filter capable of optimally trading off distortion tolerance and noise tolerance with discrimination ability. We analyze the elements used to synthesize the OTSDF filter and examine the classification ability of the OTSDF when used in a Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) system. We also introduce the Optimal Tradeoff Distance Classifier Correlation Filter (OTDCCF) which can separate two classes of targets and reject non-targets using a distance measure. A potentially fast method of implementing these correlations is with an optical correlator. We introduce the OTSDF design method that includes the physical limitations of optical correlator devices. The resulting filter can be implemented on a constrained operating curve spatial light modulator and still exhibit optimal tradeoff ability. In summary, we introduce, analyze and evaluate optimal tradeoff composite correlation filters capable of recognizing distorted targets as part of an ATR system which can be implemented either digitally or optically.