Research On Characters Extraction And Recognition Method Of Off-line Handwritten Chinese Character Based On Procedure Neural Networks

Off-line handwritten Chinese character recognition is a popular researchpoint in recognition field, and it has important theoretical significance andpractical value. Because Chinese characters’ kinds are many, structure iscomplex, similar words are many, and offline handwritten Chinese charactershave deformation but no handwriting process information. Consequently,Off-line handwritten Chinese character recognition is one of difficultproblems in Chinese character recognition field. In this paper, the theory ofprocedure neural network is used, the thinking process of human knowledge ofChinese characters is imitated, characters extraction and recognition methodof off-line handwritten Chinese character based on procedure neural networksis studied, the machine’s knowledge of offline handwritten Chinese characteris explored.The main works in this paper are as follows:1. A data table of offline handwritten Chinese character features isestablished. The writing habits and the input method of Chinese characterbarcode are imitated, Chinese characters are represented as the features of stroketype’s number, stroke type’s location, topology type’s number, the information ofstroke order. A data table of offline handwritten Chinese character features isestablished based on the study of the fault-tolerant characterization of offlinehandwritten Chinese characters.2. The model of process neural network is analyzed and established forextracting stroke characteristics of Chinese characters. Because of thecapability of dealing with space-time information of procedure neural network,the model of procedure neural network is established for extracting strokecharacteristics of Chinese characters, and improved BP algorithm is used totrain the network model. The trained process neural network is used to extract the characters of offline handwritten Chinese character. Then the data table ofoffline handwritten Chinese character features is finished.3. The fault-tolerant matching rules are made. The fault-tolerantmatching rules are made by imitating the human ability of learning andcomparison judgment. The data tables of offline handwritten Chinesecharacter features and standard Font Printed Chinese character features arematched.MATLABR2009b is selected to create simulation environment. Choosehandwritten Chinese sample characters in SCUT-IRAC HCCLIB which haveevery kind structure to validate the handwritten Chinese character recognitionsystem. Experimental results show that the method can extract characters ofoffline handwritten Chinese character, and the recognition rate is high.