Automatic visual classification of printed documents

The author presents the knowledge-based document-analysis system ANASTASIL. The system uses a formalism for document layout description and modeling. The model is realized by a tree structure, which describes the layout of a document page in different layout abstraction levels. The tree is used to initiate a best-first search in combination with a hypothesize-and-test strategy to establish a high-level electronic representation of the contents of a document. Results obtained in the different analysis phases are shown and confirm the soundness of the approach