A fast multiresolution text line and non text-line structures extraction and discrimination scheme for document image analysis

This work is devoted to the description and the evaluation of a generic method for locating text and lines in document images. One of the possible applications of this scheme is to detect and carefully extract the destination address block (DAB) on flat mail pieces. Concerning the low level design stages, the methodology is based on a multiresolution description of the document image and a joint extraction of both text lines and non text lines using their global aspect. Intermediate and high level stages consist of grouping text words into block structures, and this is performed in a efficient and robust way relying not only on some lines properties, but also on the adjacent lines