A character segmentation method for Japanese printed documents coping with touching character problems

Proposes a character segmentation method for Japanese printed documents. Since character segmentation is a kind of a search problem, avoiding `combinatorial explosion´ is essential in realizing practical systems. Segmentation is very complicated especially when characters touch each other. The method described gives a multi-stage algorithm, where the earlier stages treat more reliable segmentation than the later stages which utilize information obtained from the results of earlier stages. Segmentation hypotheses are generated in each stage on the basis of the results of earlier stages, and they are verified by the character recognition results. Experiments on more than one hundred documents have proven that this method is efficient and accurate for practical applications