Optical fuzzy computing and applications in image processing and pattern recognition

The dramatic growth in the number of fuzzy logic applications requires computing systems that can process various fuzzy logic and inference operations fast and efficiently. This dissertation focuses on developing high-performance and cost-effective optical fuzzy computing systems and applying them to selected problems of optical image processing and pattern recognition. We investigate the corresponding optical architectures, algorithms, performances and applications.; Optical fuzzy computing systems include optical fuzzy logic processing and optical fuzzy inference systems. Two optical parallel fuzzy flip-flops have been designed for fuzzy sequential information processing. The existing optical fuzzy inference systems are either digital or analog. The analog systems are costly, and perform unpopular product-sum inference and only Gaussian membership function. The digital systems, on the other hand, lack multiple fuzzy rule parallel processing capability. To overcome these problems, we propose a spatial-light-modulator-based fuzzy system that implements a more general max-min inference and various membership functions. It is easy and inexpensive to implement. A membership function decomposition technique has been developed to overlap multiple fuzzy rules without losing information. This technique can significantly reduce the optical hardware and increase data processing capability of optical fuzzy systems. For triangular-partition fuzzy systems, in particular, it has been shown that all the fuzzy rules can be superimposed into one memory matrix resulting in a simpler optical implementation.; The application of optical fuzzy computing systems to perform image distance transforms and switching median filtering has been explored. Fast image distance transforms are achieved by implementing a modified morphological algorithm using the optical fuzzy logic processing system. A new impulse detector for switching filters has been proposed to improve noise detection, which is easy to implement optically. It has been extended to a fuzzy detector for further performance improvement.; We investigated the performance of morphological correlation-based pattern recognition in terms of various standard criteria. It has been shown that morphological correlation yields better performance than linear correlation. To alleviate the computational complexity of the morphological correlation, a simplified scheme has been proposed. The fuzzy postprocessing in pattern recognition is enhanced by using a modified fuzzy-rule-based method.