Research Of Handwritten Chinese Characters Multi-Classification Recognition Based On Structure Features Of Characters

Off-line handwritten Chinese character set has features of large quantity,complex structure,many similar characters and severely ruleless distortion, which makes that off-line handwritten Chinese character recognition becomes the largest problem and one of the final goals in the domain of character recognition. But that human ocular apperceive is truly a robust character recognition system with the ability of much fault-tolerance which can adapts to all kinds of noises. Recent years, much research on humanoid recognition of off-line handwritten Chinese characters has been made, but it is still a significative research on how to improve the agility of computer humanoid recognition on off-line handwritten Chinese character in feature's use and recognition methods.In this paper, some research and some useful conclusions are made based on seriously studying and summarizing present abroad adopted recognition methods and former's work.The following are the major contents in the thesis:1,The multimode qualitative recognition of handwritten Chinese character image. A classified method of Chinese character's structure and component's complexity are presented. The type code and decomposing arithmetic of Chinese character's structure are given, and the extracting arithmetic of Chinese character's complexity classification is given. The classification and extraction of handwritten Chinese character's structure and component complex degree are realized. The result of experiment shows it is doable.2,The system design of humanoid Chinese character recognition based on the characteristic of the Chinese character structure. This system adopts a two-level paralleling structure of decision-making level and pattern recognition level. Decision-making level chooses the best recognizing strategy and parameter by the recognized Chinese model, Pattern recognition level uses recognizing strategy to do multi-classifier matching recognition, then the recognition efficiency of large sort Chinese characters and the practicability of Chinese character recognition system are improved.3,The theory introduction and characteristic analysis of handwritten Chinese character meticulous classification methods. From the beginning of statistic and structure pattern recognition methods, the former research results of my lab are analyzed such as the wavelet gridding method,the biomimetic pattern recognition based on double-weights Elliptical Neural Vector,eight code method and the stroke segment extracting method based on procedure neuron. Based on their recognition characteristics, the best recognizing strategy and parameter are choosed to realize multi-classifier matching recognition.4,Application of multilevel and multistage recognition system. Handwritten Chinese characters in SCUT-IRAC are selected and the VC++ and the MATLAB6.5 are used as the realization methods in this thesis. The recognitions of handwritten Chinese character images with various types are accomplished successfully.The research in this thesis shows that: the multimode recognition system based on decomposition of Characters' Structure can make coarse classification, the best recognition strategies decision-making and multi-classifier results matching recognition. The experiment shows that the process of human recognition based on the whole and resolvability of handwritten Chinese character complexity and structure can be imitated by this method.