Optical Adaptive Pattern Recognition

Optical pattern recognition has long been considered a very useful tool in machine vision, robotics, automation and image understanding. Adaptive pattern recognition is to recognize patterns through iterative procedures. The key factors for optical adaptive pattern recognition are to identify feedback parameters, to devise algorithms to adaptively change the system's parameters according to the feedback signals. This dissertation is intended to advance the investigation of the role that an optical correlator based on the joint transform architecture can play in adaptive pattern recognition process. In this research work, algorithms for an adaptive joint transform correlator are developed and tested under full scope of computer simulations. This research also presents a new algorithm for adaptive learning: two step adaptive learning. In the feedback process of this algorithm, the saturating phenomenon in the intensity of correlation peaks and the average rejection ratio between interclasses in the training set are used as the guidelines in designing a proper feedback scheme. Investigations of the relationship between input patterns and their correlation distribution are also presented. These investigations indicate that: not only correlation peaks, but also correlation distribution can provide us informations about input patterns. A case study of utilizing the symmetric properties of correlation distribution in classifying input objects is also introduced.