A step towards understanding paper documents

Abstract

This report focuses on analysis steps necessary for a paper document processing. It is divided in three major parts: a document image preprocessing, a knowledge-based geometric classification of the image, and a expectation-driven text recognition. It first illustrates the several low level image processing procedures providing the physical document structure of a scanned document image. Furthermore, it describes a knowledge-based approach, developed for the identification of logical objects (e.g., sender or the footnote of a letter) in a document image. The logical identifiers provide a context-restricted consideration of the containing text. While using specific logical dictionaries, a expectation-driven text recognition is possible to identify text parts of specific interest. The system has been implemented for the analysis of single-sided business letters in Common Lisp on a SUN 3/60 Workstation. It is running for a large population of different letters. The report also illustrates and discusses examples of typical results obtained by the system.