Hybred: An OCR Document Representation for Classification Tasks

The classification of digital documents is a complex task in a document analysis flow. The amount of documents resulting from the OCR retro-conversion (optical character recognition) makes the classification task harder. In the literature, different features are used to improve the classification quality. In this paper, we evaluate various features on OCRed and non OCRed documents. Thanks to this evaluation, we propose the HYBRED (HYBrid REpresentation of Documents) approach which combines different features in a single relevant representation. The experiments conducted on real data show the interest of this approach.