Segmentation of document images based on texture features

This paper presents a method for typography segmentation at a text block level in document images using generic adapted image processing tools. Our proposition is based on the concept of texture, assuming that text can be seen as a juxtaposition of textured regions, each one corresponding to a particular typography. Thus, we show on a few examples that texture modeling and segmentation tools (namely Gabor filters and the geodesic active regions model) can be adapted to locate the typographical zones composing a document image. Finally, we can say that our method is generic enough to cope with a lot of recurrent problems in document (e.g. noise, skewing,...). This paper is proposed as an extension to a preceding study on the use of Gabor filters at a character level.