Research On Handwritten Chinese Character Recognition Experimental Platform And Stroken Based Elastic Meshing Method Of Feature Extraction

The recognition of handwritten Chinese character is a comprehensive research topic.The research not only has its theoretical value, but also has its practical value. Because ofthe difficulties of Chinese characters, such as the big number of Chinese characters, thedifferent structures, many similar characters and distortion of the font style, the researchbecomes one of the most challenging topics in the field of pattern recognition.The establishment of a unified recognition experiment platform for the handwrittencharacters is an important way to evaluate objectively and analyze the various featureextraction and classification algorithms.In this paper, we designed and constructed an experiment platform for the research.Based on it, different algorithms were applied and studied. The mainly contents of thispaper are:1. The design and research of recognition experiment platform which is for Chinesecharacters. In this paper, we designed and constructed a sample database of handwrittenChinese characters. Then, some preprocessing techniques which were commonly used inChinese character recognition were introduced. After that, design and construction of theexperiment platform was focused on. Then, the implementation of experimental platformwas introduced.2. Research on the feature extraction methods. Feature extraction method has twocategories: the method which based on the structure feature and the method which based onstatistical features. In this paper, they were studied. Then, we proposed a stroke-densityfunction based double elastic meshing method, which can extract the feature of left-fallingand right-falling strokes effectively in a Chinese character image.3. Research on the classification. Firstly, some typical classification methods wereintroduced, including K Nearest Neighbor (KNN), Bayes discriminated, Support VectorMachine (SVM) and BP neural network algorithm. Then, the paper presented the theory ofAdaBoost algorithm. After that, an AdaBoost method which was based on BP network forthe classification was studied.4. Experimental results and analysis of characters recognition. Firstly, analyzing theexperimental results of running time to divide and construct the sample sets by the platform. The practicability and efficiency of the platform which was built by this paper were verified.Then, based on the platform of character recognition, extraction ways of different featureswere experimented, and results have verified the effectiveness of the new feature extractionmethod which was proposed by this paper. At last, different classification methods wereexperimented, and the results show that the AdaBoost means which was based on thenetwork of BP has a high correct rate than others.