Abstract

An elementary binary pattern classifier has n binary input variables and one binary output variable. The input variables may be identified with image pixels so that there are 2n possible input patterns. Classification or specification of the binary output is required for m of these patterns; the remaining patterns are do not cares. For m > 2n and large n (e.g., n ≥ 64), classifiers using all-electronic combinatorial logic techniques generally require numerous decision elements and complex interconnections. These requirements may lead to limited performance in areas such as processing speed and noise tolerance. Classifiers based on holographic associative memory techniques and the use of optical phase conjugation have potential for providing substantial performance improvements in these areas. In particular, analytical and computer simulation results indicate that phase conjugation with thresholded gain may significantly impove the ability of holographic associative memory processors to correctly classify numerous noisy or highly cross-correlated binary patterns. This indication may be attributed mainly to the fact that an additional well-interconnected decision surface can be inserted in a classifier by a thresholded phase conjugator.

© 1986 Optical Society of America