Study Of Handwritten Chinese Characters Recognition Based On Gabor Double Elastic Mesh Feature Extraction

With the unceasing enhancement of science and technology level, the off-line handwritten Chinese character recognition technology has become one of the widely used technology in industries. Because Chinese character is composed of a horizontal stroke, a vertical stroke, a right-falling stroke and a left-falling, so how to accurately and quickly identify the features of Chinese character strokes which has become the difficulties of the study of handwritten Chinese characters recognition. The method of feature extraction is one of the most important steps of Chinese characters recognition. Effective feature extraction method can accurately and quickly recognize the features of Chinese character strokes. The main contents of this paper are as follows:1. We collect some samples of handwritten Chinese characters, and then introduce several commonly used methods about image preprocessing. Based on the handwritten Chinese characters image preprocessing, the image of Chinese characters becomes more easy to be recognized.2. Research on the feature extraction methods of handwritten Chinese characters. This paper introduces several commonly used feature extraction methods about recognition of Chinese characters. Aiming at the situation that the characteristics of the image of handwritten Chinese characters extracted by Gabor transform can suppress the image interference such as noise and fuzzy of the Chinese characters, but the image of the Chinese characters which the text appears deformation because of writing in different ways can not be identified effectively, a new method of double elastic mesh feature extraction based on Gabor is proposed by combining the double elastic mesh technology and the Gabor feature extraction approach improved. With this approach,firstly, a handwritten Chinese character’s image is put into a set of horizontal, vertical, right-falling and left-falling directional stroke images. And then the image features of the four directions are extracted through the obtained double elastic meshes and the Gabor filter group which is set with the optimal parameters. The experiments dealing with lots of handwritten Chinese characters prove that the calculation is reduced more significantly and the recognition speed is improved more significantly than the other feather extraction methods. This approach is illustrated that can effectively avoid text deformation and noise problems. And the recognition rate is improved effectively.3. Research on the classification methods of handwritten Chinese characters. This paper mainly introduces several common classification methods which are nearest neighbor classification(KNN), naive bayesian classification, support vector machine(SVM) and BP neural network algorithm. Then we compare the effectiveness of the several kinds of classification methods.4. Experimental results and analysis of handwritten Chinese characters recognition. Handwritten Chinese characters are did contrast experiments by using several common methods of feature extraction and methods of classification recognition. And the experimental results are analyzed. The experimental results indicate that the method of double elastic mesh feature extraction based on Gabor is effective. And when the samples are classified and recognized by BP neural network classifier, the correct rate of recognition is the highest.