Studies on Development of OCR system for Historical Documents and Application to Technologies in Electronic Dictionary

本研究之目的在于建立史学、古文字、文学专家以手提与笔记两种体裁，对历史手抄文献进行电脑解译的电子词典“久久寺Kaidoku词典”与“久久寺游礼词典”，并开发可在手提环境下使用的电脑化词典。此外，将词典直接应用于上述历史文献音译支持系统（historical document OCR）中的字符识别研究。本研究期间获得的研究成果如下：(1)将作为“Kuzushiji Yourei字典”索引的图像（它允许我们基于笔画（Kihitsu-jun）索引检索字母的形状和字母使用示例）输入为具有“Kuzushiji Yourei字典代码”、“Mojikyo代码”和“Shift-JIS”内部代码等属性的图像。(2)开发了在上述词典中搜索相似字符的检索功能(3)将“n-gram”方法应用于历史文献音译支持系统（历史文献OCR）的研究。(4)为在该历史文献上建立约24万个字符的字符模式字典，用于识别过程，开发了切分程序并进行了字符选择工作。(5)将历史文献字符数据库中的HCD系列第二版作为计算机化词典之一制作完成。(a) HCD2，债务债券标题行，府宫禅北文件，200行，1378个字符，二进制格式。(b) HCD2a，债券标题行，府宫禅北文献，200行，1378字，256步。c HCD2b，债务债券标题行，府宫禅北文件，200行，24位1378个字符的彩色格式。(d) HCD3，债务债券标题行，府宫禅北文档，183个字符类型，4933个字符，二进制格式。(6)文档中以标题行为重点的字符识别使用上述字典进行。对标题行中每个字符不分割匹配字符模式的识别技术进行了研究。(7)对词典制作的《久久纪Kaidoku词典数据库》中提取的笔画顺序进行了估计研究。本研究的研究报告，包括中间版本，分别发表于2001年3月和2000年3月，此外还有关于历史文献音译辅助系统Less的论文

The purpose of this research is to build the electronic dictionaries, "Kuzushiji Kaidoku dictionary" and "Kuzushiji Yourei dictionary", used in which the specialist of the historical study, paleography, and literature deciphers the historical handwritten documents using the computer including mobil and note book styles, and to develop the computerized dictionary that can be used in a mobil environment.Moreover, it is to apply the dictionaries directly to the character recognition researches in the transliteration supporting system for historical documents (Historical document OCR) mentioned above.The following research results were obtained during this reserch period.(1) The images which is the index of "Kuzushiji Yourei dictionary" (it allows us to retrieve the shape of letters and examples of letter use based on the stroke (Kihitsu-jun) index) were input as the images with attributes such as "Kuzushiji Yourei dictionary code", "Mojikyo code" and "Shift-JIS" internal code, and an elec … More tronic Moji database was built(2) A retrieval function which the user can search the similar characters in the above-mentioned dictionary was developed(3) The "n-gram" method was applied to the researches in the historical document transliteration supporting system (historical document OCR), and it was confirmed that "n-gram" was effective when the lost or missing charahter in the document was presumed(4) To build the character pattern dictionary of about 240,000 characters on the historical document to be used in the recognition process, a development of segmentation program and the character selection work were carried out(5) The second edition of HCD series below in the historical document character database had been made as one of computerized dictionaries. (a) HCD2, title line for debt bond, Fushimiya Zenbei document, 200 lines, 1,378 characters, and binary format. (b) HCD2a, title line for the bond, Fushimiya Zenbei document, 200 lines, 1,378 characters, and 256 steps. c HCD2b, title line for debt bond, Fushimiya Zenbei document, 200 lines, and 24bits 1,378 character colors format. (d) HCD3, title line for debt bond, Fushimiya Zenbei document, 183 character types, 4933 characters, and binary format(6) The character recognition in the document focused on the title line was carried out using the above-mentioned dictionary. The research of the recognition techniques for matching the character pattern without segmentation for each character in title line was developed(7) Study on estimation for stroke order extracted from "Database of Kuzushiji Kaidoku dictionary" made by the dictionary has been carried out. Research reports including intermediate version for this study were published in March, 2001 and 2000 respectively besides papers regarding the historical document transliteration supporting system Less