Research Of Humanoid Recognition On Off-line Handwritten Chinese Characters Based On Procedure Neuron

In recent years, because of commerciality and practicability of off-line handwritten Chinese character recognition, the exploration in this area has been a hot topic in research and application of pattern recognition. However, AI (artificial intelligence) and automatic recognition are just beginning, and many conceptions and theories are needed to be improved. Overall, off-line handwritten Chinese character recognition is very difficult. At present, there is no unitary systematical description of definition, theory and method on it, and the solid and systematical theory fundament is needed. But we may predict that with the development of theory of system, and AI and computer technology, off-line handwritten Chinese character recognition will be improved a lot, and will be used widely in practical.The following are the major contents in the thesis:1,Analysis of mechanism of humanoid recognition on off-line handwritten Chinese character and extraction of basic stroke segment. The process of Chinese characters "cognizing" by human is analyzed according to the relative changeless topological structure of compounding, connection and crossover of some basic Chinese character strokes; Procedure neural sequences are used to extract basic strokes segments through space-aggregation and time-aggregation, and the influence of character distortion on stroke segment cognition has been eliminated radically.2,Extraction of Chinese character strokes and identification of joint and intersection point. The psychological process of stroke styles fault tolerance and identification of topological structure are analyzed in Chinese characters "cognizing" by human. Six styles of Chinese characters strokes with fault tolerance are concluded; Composing rules of Chinese characters strokes and the identification rules of joint and intersection styles are defined. Chinese characters strokes with fault tolerance and their joint and intersection styles could be cognized by simulated human beings, then Chinese characters humanoid recognition could be realized.3,Construction of Chinese characters multi-characteristic knowledge data-base table and humanoid recognition with fault tolerance. Chinese characters recognition process of learning, memorizing, contrasting and judging by human are imitated, the multi-characteristic knowledge data-base table is built, which aims to the best cover of Chinese characters strokes in Chinese characters image; And samples of Chinese characters are built, and humanoid recognition rules with fault tolerance are defined.4,Experiment and analysis of the results. 100 different handwritten Chinese characters in SCUT-IRAC are tested. The method is proved to be accurate and fault-tolerant. However, in order to improve the fault-tolerant ability, there are still some Chinese characters denied or inaccurate to be recognized.