Improving speed of Japanese OCR through linguistic preprocessing

The author proposes uses of linguistic and statistical knowledge in preprocessing in character recognition to achieve higher speed processing. In this approach, character prediction substitutes for rough classification which occupies a large part of the recognition time. Three character prediction methods are proposed and prediction capability is verified. This method is implemented in a recognition system to evaluate its performance. Accuracy is compared with the result of the postprocessing error correction method. As a result, it is observed that the predictive recognition method has a considerable speed advantage. For almost the same recognition accuracy, this method is 20% faster