Modelisation and Classification for Retrospective Conversion of Documents

International audienceThis article presents a document representation oriented towards retrospective conversion. Retrospective conversion of documents allows to convert document images into an electronic format. Classification methods using the proposed representation are presented. Some of these classification methods are used to classify the constituant elements of documents. Others are used to classify whole documents. These classification methods are combined into a global strategy for retrospective conversion. This strategy is based on a cycle where different understanding levels interact. This so called " understanding cycle " is designed around the notion of inconsistency detection and solving. Some results concerning object classification as well as document classification are presented. These results emphasize the influence of the interpretation cycle in the retrospective converision process