A Chinese OCR System Based On Gabor Features And SVM

The Chinese character recognition system is a system which extracts the text information from an image. According to different types of images, it has several applications in a wide field. For example, the bank bills recognition, the digital library, postal code classifier, handwriting recognition and so on.Since the computer's born, the character recognition algorithm is the hotspot for science research for a long time. As a whole, there are two main methods: one is based on statistic while the other is based on rules.In the statistic way, people use image features to classify it. After many years of study in this subject, Gabor features become the most important features for material, face and character recognition. As a wavelet method, Gabor features performs similar principles like human visions, which has good biology background. So it can solve recognition problems very well.As to the classifier, the Support Vector Machine (SVM) which is appeared in 1990s has gained much attention in this field, and become the most active subject. Since SVM has good effect in the situation that training samples are not enough, it is one best choice for OCR system.This paper studies both two tools and learns how the correctness rate and recognition rate change in different situations. In the basic framework of Chinese OCR system, we use binary cluster algorithm to find the text regions in an image. Then we integrate Gabor features and SVM in OCR system, and develop a demo system for experiment. The correctness rate and recognition rate raise up though a great deal of training, and the test proves that Gabor features and SVM archives satisfactory result.