Developing a Historical Document Analysis System Using Hand-written OCR Techomology

在本研究项目中，我们可以得到以下结果。(1)历史文字识别的基础研究。我们研究了历史字符的字符识别和切分的一些特殊要求。利用日文手写字符识别技术，在历史字符的小数据集上，获得了95%的正确识别率。我们还开发了一些新的历史字符切分和归一化技术。(2)用于识别研究的历史字库。我们建立了一些数据库，对历史文字识别的研究进行了推进。该数据库包含超过25万个字符。(3)历史文献分析系统的接口原型。开发了历史文献分析系统的原型系统。一种是具有使用n元语法信息呈现不可读字符的正确字符候选的功能的系统。另一个是历史文字电子词典原型系统。该词典具有相似字搜索功能，是手写字符识别的应用。

In this research project, we could obtain the following results. (1) The basic research on the Historical Character Recognition. We examined on some special requirements for the character recognition and segmentation of historical characters. By using a small data set of historical characters, we obtained 95 percents of correct recognition ratio applying Japanese hand-written character recognition technology. We also developed some new technology of segmentation and normalization of historical characters. (2) Historical character database for recognition studies. We made some database to put the study of historical character recognition into progress. The database contain over 250 thousands characters. (3) Interface prototyping of historical document analyzing system. We developed prototype systems of historical document analyzing system. One is a system with a function of presenting correct character candidates of unreadable character using n-gram information. The other is a prototype system of electrical dictionary for historical characters. The dictionary has a function of similar character search, which is an application of hand-written character recognition.