Research Of Recognition On Off-line Handwritten Chinese Characters Based On Procedure Neural Networks

Due to many complex structures of Chinese characters and diverse distortion of handwritten Chinese characters, Off-line handwritten Chinese character recognition is one of difficult points in Chinese character recognition field. Facing the difficult point of Off-line handwritten Chinese character recognition and setting about from simulating the human vision character, procedure neural networks is used and machine vision Off-line handwritten Chinese character with fault tolerance is proposed in this thesis.The following are the major contents in the thesis:1. A method for extracting etymon of Off-line handwritten Chinese character is defined. The process of Chinese characters"cognizing"by human is imitated according to the some etymon; a method of using procedure neural networks to extract etymon of Off-line handwritten Chinese character through space-aggregation and time-aggregation is defined, and the influence of character distortion on strokes cognition has been eliminated better.2. The establishment of etymon and Construction of Chinese characters multi-characteristic knowledge data-base table for Off-line Chinese character. Some research on Chinese character structure is done and trait of computer vision is considered.Based on the summarization of some coding schemes of Chinese character, nineteen kinds of etymon are determined and a new multi-characteristic knowledge data-base table is built; Sample knowledge bases of sample Chinese characters are built.3.Training and testing of procedure neural networks.Some Off-line handwritten Chinese characters in SCUT-IRAC are selected to be training sample of Procedure neural networks.The Pre-defined etymon are extracted through space-aggregation and time-aggregation.Procedure neural networks are trained and tested with using Pre-defined multi-characteristic knowledge data-base table of Off-line handwritten Chinese character.The network structure are determined and the appropriate network parameters are saved.4.Humanoid recognition with fault tolerance of Off-line handwritten Chinese character.Chinese characters recognition process of learning, memorizing, contrasting and judging by human are imitated.The humanoid recognition rules with fault tolerance are defined and Off-line handwritten Chinese characters are recognised.Based on the proposed method of Off-line handwritten Chinese characters recognition, 300 different Off-line handwritten Chinese characters in SCUT-IRAC are tested. The experiment results proved the feasibility and validity of the method.