Information extraction from document images using white space and graphics analysis
The goal of this work is the fast extraction of relevant information from document images. Examples of interesting information are the type of document (e.g. form, report, letter), the title of an article or the sender of a business letter, and a logo or figure on a page. The basic idea is to use non-textual cues from the document image before any OCR/ICR or word recognition is performed. The approach is based on a compact runlength representation of the binary image and allows a document type classification by white space analysis in a time comparable with the input of the compressed image. Graphics related information extraction needs approximately the same time.
- Gerd Maderlechner, Symbolic Subtraction of Fixed Formatted Graphics and Text from Filled in Forms, Proc. IAPR Workshop on Machine Vision and Applications, Tokyo, Nov. 1990, pp. 457–459.Google Scholar
- M. Ozaki, Logical Tagging of Document Images by White Space Pattern Matching, in Shape, Structure and Pattern Recognition by D. Dori and A. Bruckstein (editors), World Scientific, Singapore, 1995, pp. 350–359.Google Scholar
- D. Rus and K. Summers, Using Non-Textual Cues for Electronic Document Browsing, in Digital Libraries: Current Issues by N. R. Adam, K. Bhargava, and Y. Yesha (editors), Lecture Notes in Computer Science, Springer Verlag Berlin, New York 1995, pp. 129–162.Google Scholar
- P. Suda, C. Bridoux, B. Kämmerer, G. Maderlechner, Logo and word matching using a general approach to signal registration, Proc. ICDAR'97, pp. 61–65.Google Scholar