A Novel Segmentation and Recognition Algorithm for Chinese Handwritten Address Character Strings

This paper presents a new method for segmenting and recognizing Chinese handwritten address character strings. First, a dissection algorithm is applied to over-segment string image into radical series so that the correct segmentation could be achieved by merging those radicals according to the correct merging path. Then, the method synthesizes layout analysis, isolated character classifier and bi-gram language model to find the best merging path and the best recognition result. The classifier used in this paper will give every isolated character image ten recognition candidates, each with corresponding recognition confidence. The parameter of bi-gram model is obtained from address database which contains more than one hundred thousand address items. In experiments on 946 mail images, the proposed method achieves correct rate of 87.2 percent