Visual Perception of Unitary Elements for Layout Analysis of Unconstrained Documents in Heterogeneous Databases

摘要

The document layout analysis is a complex task in the context of heterogeneous documents. It is still a challenging problem. In this paper, we present our contribution for the layout analysis competition of the international Maurdor Campaign. Our method is based on a grammatical description of the content of elements. It consists in iteratively finding and then removing the most structuring elements of documents. This method is based on notions of perceptive vision: a combination of points of view of the document, and the analysis of salient contents. Our description is generic enough to deal with a very wide range of heterogeneous documents. This method obtained the second place in Run 2 of Maurdor Campaign (on 1000 documents), and the best results in terms of pixel labeling for text blocs and graphic regions.