A Study on Design and Implementation of Image-Based Scholarly Information System

近年来信息处理技术尤其是图像分析技术的进步，使得学术信息在传统文本的基础上，也可以通过图像进行积累和传播。本研究旨在建立以文献影像为表征之学术资讯的分析、储存与利用方法，以改善学术资讯资料库的建构与人机介面。在文档图像分析方面，我们开发了一种对文档图像进行分析并自动构建数据库的方法。该方法是基于概率语法的，概率语法是一种用于处理二维空间中布局的对象的扩展语法。在文档图像存储方面，我们研究了一种近似字符串匹配的方法，从现有图像分析方法得到的错误文本中提取信息。在利用方面，我们开发了一个实验性的书目整合系统，将目录、文章和参考文献中的书目数据整合在一起。通过该系统，对所开发的文献图像分析与存储方法的可行性进行了评价，证实了该方法的有效性，可以应用于书目集成。

Rescent progress of information processing technology especially on the image analysis enables to accumulate and distribute scholarly information with images as well as conventional texts. This research project aims to establish a methodology for analyzing, storing and utilizing scholarly information represented with document images for improving the scholarly information database compilation and the human machine interface. As for the document image analysis, we developed a method for analyzing document images and construct a database from the document images automatically. The developed method is based on the probabilistic grammar, which is an extended grammar for handling objects layouted in two dimensional space. As for the document image store, we studied an approximate string matchig method to extract information from erroneous texts which is obtained by the developed image analysis methods. As for the utilization, we developed an experiemental bibliographic integration system that integrates bibliographic data appearing in table of contents, articles and references. By this system, we evaluated the feasibility of the developed document image analysis and storage methods and confirmed that the method is effective enough to apply to the bibliographic integration.