Résumé

Le but de cet article est de proposer une méthode pour la séparation entre manuscrit et imprimé dans des documents. La méthode proposée repose sur des descripteurs originaux appartenant

Abstract

In this paper, we address the issue of separating handwriting from printed text in doc- ument images. We present a reliable method based on a novel set of features belonging to two different categories, linearity and regularity, invariant to translation and scaling. Specifically, a linearity measure derived from the histogram of segment lengths is introduced. The resulting framework is independant of the document layout, supports any latin langage used and provides a computationally efficient technique. Its performances, assessed on real documents, allow a recognition rate that exceeds 90%.