A study of the character recognition system for cursive script in historical documents

点击翻译按钮获取中文摘要

In this research we have studied some recognition systems of continuous cursive scripts (Hiragana) by the conventional pattern recognition methods such as a k-NN rule and an artificial neural network. Also we have analyzed the characteristics of these continuous cursive scripts in historical documents to improve the character recognition performance. Moreover the one of comical and popular books published in Edo period in Japan, which was called as 'KI-BYOSHI', was examined, and then reported the frequency of scripts. The results of this research are as follows :1. The continuous cursive scripts data including each pixel information and category name were prepared.2. The character recognition system for the continuous cursive scripts was developed and the difficulty for correct recognition due to the continuous writing was discussed.3. The character recognition system based on the order of writing or writing procedure was proposed, and its performance of recognition rate was evaluated.4. The statistical analysis for the frequency of each script in the several historical documents in various periods was executed, then it was shown that the frequency of Hiragana had increased as the period became newer.