“…These methods can lead to better results on the heterogeneous documents but they require a lot of a priori knowledge. The literature presents yet more recent approaches, which are more robust and less dependent on a priori knowledge, that are mainly based on the characterization of the document texture [16,17], the multi-scale analysis [18], the edge analysis [19],the grammatical model learning [20], the rules-based techniques [21] or the stochastic methods [22].Consequently, most of the proposed methods either require a priori knowledge related to the nature or the document structure, a high computing time and enormous resources, or they are not suitable for documents with potential defects and damages. In addition, most of the suggested works does not lead to an evaluation of their performances or to a comparison with other approaches.…”