A Study on Features for Improving Performance of Chinese OCR by Machine Learning

This paper discusses a method to improve the performance of Chinese OCR by choosing a proper feature vector and synthetic classification. We compare two groups of features which are used to implement Chinese OCR System and demonstrate that the first group of features is more useful for static Chinese OCR System. By now feature extractions have been done either for local features or for global features. Classifications have been done by single classification. We propose synthetic features extraction and classification in this paper. We find that the result is improved by machine learning method. Later we apply the result in the area of off- and on-line signature verification system.