Handwriting analysis: segmentation and recognition

The Document Analysis and Recognition research activity at Telecom Paris is oriented towards two goals. The first one is the segmentation of handwritten pages of document into text lines. The second one is the recognition of handwritten cursive words using a knowledge based strategy. At present the two stages are not combined as the segmentation into words is not performed. We have developed a new method for segmenting handwritten pages of documents into text lines. This method aims to be general, i.e. dealing with unconstrained documents such as address blocks, rough drafts and letters. In such documents, contrary to printed documents, specific problems arise such as different line directions appearing within a page, lines overlapping or being interwoven. The method is an iterative process which relies on perceptive mechanisms which enable human beings to see text lines without reading the document.