Pi_{ODA} : the paper interface to ODA

Abstract

In the past, many people have proclaimed the vision of the paperless office, but today offices consume more paper documents than ever before. As computer technology becomes more and more important in daily practice of modern offices, intelligent systems bridging the gap between printed documents and electronic ones, called paper-computer-interfaces, are required. In this report our model-based document analysis system Pi_{ODA} is discussed in detail. Basic ideas of the ODA standard for electronic representation of office documents are the foundation of our document model. Moreover, different knowledge sources essential for the analysis of business letters are incorporated into the Pi_{ODA} model. The system comprises all important analysis tasks. Initially, layout extraction includes a necessary low-level image processing and segmentation to investigate the layout structure of a given document. While logical labeling identifies the logical structure of a business letter, text recognition explores the captured text of logical objects in an expectation-driven manner. By this way, word hypotheses are generated and verified using a dictionary. Finally, a partial text analysis component syntactically checks well-structured text objects, primarily the recipient of a letter. As output, Pi_{ODA} produces an ODA conforming symbolic representation of a document originally being captured on paper. Now, the document is available for any further automatic processing such as filing, retrieval or distribution. The inherent modularity of our system, however, allows a reuse of knowledge sources and constituents of the architecture in other document classes such as forms or cheques. Additionally, Pi_{ODA} is an open and flexible system: improved and new analysis methods can be integrated easy without modifying the overall system architecture.