A large scale Japanese handwritten address recognition system using rough and fine classification

Post-preprocessing is an important part in handwritten character recognition system. In post-preprocessing, the knowledge about address is used to verify the result of segmentation and recognition. As a result, the best-matched output of segmentation and recognition is selected. The main problem in post-preprocessing is there are too many addresses to be handled, so that the processing time is tremendously long. In this paper, we propose a post-processing technique with rough and fine classification. Especially in rough classification, we introduce the priority allotting method(PAM) for selecting address candidates with accuracy and speed. We show the effectiveness of this method and the recognition system constructed by applying the proposed method.