A Strategy for Retrospective Conversion of Documents

International audienceThis paper proposes a strategy for retrospective conversion of documents. This strategy consists in an interpretation cycle where document analysis and document understanding interact. This cycle is initialized by the extraction of the outline of the layout and logical structures of the document. Then, each iteration of the cycle consists in the detection of inconsistencies in the document modeling. Treatment are applied to correct these inconsistencies. The cycle ends when no more inconsistency occurs. The method used for document objects classification is based on the fusion of statistical and structural clas-sifier results. A structural comparison method is used for document structure classification. The different levels of inconsistency which may occur in the document representation are detailed. Several treatments to solve theses inconsistencies are presented. Finally, future prospects concerning incremental learning of new classes of documents are developped in order to implant this strategy in a document processing system