Research On Off-line Handwritten Character Recognition Based On Support Vector Machine

Characters are very important tools in human information communication. With the rapid progress of the information technology, the demands for the recognition of handwritten characters are wider and wider in many fields, so we should enhance the research of the recognition of handwritten characters. After long term work, the research of off-line handwritten characters are gained much attention and great achievents have been gained on characters segmentation and preprocessing.The technology has been more and more perfect. However, in the aspect of recognition, the information of off-line handwritten character is only 2D pixel matrix, and the handwritten characters have so many shape distortions, so it is difficult to recognize the hand written characters effectively. Besides, traditional statistic mode for recognition is under the condition of adequate samples.That is, fairly effect could be obtaineed only if the number of samples runs to infinite.But in fact, the number of samples is limited, and then ideal results can hardly be got on the basis of the existing methods. However, Statistical Learning Theory, which is specially designed for the case of small samples, supplied a better theoretical frame to the research of Statistical Pattern Recognition under the circumstances of limited number of samples, and presented a new pattern recognition method----Support Vector Machine(SVM). SVM is a highly adaptive method, which is applied in the areas of Pattern Recognition, Regression Estimation, Function Approximation and Density Estimation.The object of the research is off-line handwritten numerals which are written by some specific persons(including 10 Arabic numerals). Samples are Established by ourself according to the requeriment of the experiment.Some simulation researches are done with VC++6.0. The process from the characters input into the computer recognition is completed in the experiment. In the step of inputting, the inputting of characters is implemented by draging mouse. In the step of pre-processing, the preprocessing including smooth, sharpening and thinning is completed, so the next feature extraction and recognition become easy. In the step of feature extraction, a simple stencil-plate method is used to gain the feature effectively. In the step of recognition, the SVM method is used to recognize the off-line handwritten characters which are written by some specific persons. In the step of outputting, the outputting of characters is implemented by the contents of pop-up dialog box.After comparing the impact with different penalty parameters, kernel functions and kernel parameters, a good method is found to recognize the off-line handwritten characters which are wrote by some specific persons. In order to show the advantages and disadvantages of SVM method, some other usual methods are also implemented. Then the comparison and analysis are completed. The result proves that the SVM method is better than others in evidence.