Research on the digitization system of scientific documents

点击翻译按钮获取中文摘要

1.Throughout the research period, we build a ground-truthed database of page images of mathematical articles. Using the database, we developed and improved the math symbol recognition engine and the segmentation method of text areas and math expression areas. A part of the database is now open to public on our web site.2.To improve the math structure analysis method base on virtual link network developed in the previous research, we adjusted the cost of the links of the network in detail using the database above. On the other hand, we introduced a notion of "center band", calculated robustly against mis-recognition of characters, to stabilize considerably the structure analysis of math expressions.3.We developed a method to segment touched characters in math expressions using the matching of sub-patterns with other non-touched characters patterns in the same page. We also extended a framework used frequently to segment characters in text areas in a way adapted to math formulae images4.We developed a method to recognize complicated matrices including repeat symbols or area symbols, using variable block pattern elements.5.We investigated the method to detect bibliographic data and logical structure of math papers from the recognition results.6.We finally studied the recognition of commutative diagrams in math papers and graphs of elementary functions in the figures of math texts as well. These are however still on the state of trial research.7.A math document recognition software "Infty Reader" developed using the results of this research is available freely from the web site : http://www.inftyproject.org./