Collaborative Research And Application Of Pattern Recognition Algorithms

In the late of 1980's, Professor Haken, the founder of Synergetics, first described a novel approach to pattern recognition, synergetic pattern recognition. This thesis is concerned with the algorithm and its application. Firstly, it introduces the theory of synergetics and explains some basic concepts in detail. Secondly, it expounds several models of synergetic pattern recognition and their corresponding algorithm frameworks. Thirdly, it discusses in a systematic way the details and problems in the algorithm, including the selecting of prototype patterns, the setting of attention parameters and the realizing of spatial invariant pattern recognition. For the selecting of prototype patterns, the thesis presents and implements an algorithm based on genetic algorithm to train the free parameter in the learning algorithm based on the superposition of information. The experimental results show that it converges fairly fast. For the setting of attention parameters, it improves the algorithm of training attention parameters based on award-penalty learning mechanism. In addition, it studies the explicit learning parameter sub-algorithm in SCAPAP on the condition that a zero-error solution doesn't exist. Taking the classification of hepatitis and kidney globule color pathology images as an example, it conducts an experiment and concludes that SCAPAP can't always predict the final results correctly. For spatial invariant pattern recognition, according to the position invariance in Fourier Transform and the scale invariance in Mellin Transform, it proposes an algorithm based on Fourier Transform and Mellin Transform to realize the spatial invariance in synergetic neural networks. Finally, it describes the design and implementation of Pathology Image Classification System based on synergetic pattern recognition algorithm at length. The system is rather robust. Furthermore, to learn the test images which have been refused to be classified, it puts forward an information superposition algorithm based on genetic algorithm. Thus, by self-learning, the system has stronger recognition capability.