Classification of Multi-structured Documents: A Comparison Based on Media Image

摘要

This paper focuses on the structural comparison of multimedia documents. Most of the systems treating the multimedia documents exploit only the text part of these documents. However, the text is no longer the only means to carry information. The major issue is to extend these systems to the other modality notably to the image that constitutes one of the basic components of multimedia documents. The complexity of multimedia documents, multi-structured in essence, imposes not only a structural representation in the form of trees, but rather in the form of graphs. The graphs are in appropriateness to the description of these documents. For example, one will be able to describe the components of a scene of an image, the relations between these components, their positions (spatial relations), etc. We propose a new similarity measure of graphs, based on a univocal matching between the graphs to compare. In our approach, we will take account of structural information and specificities of multimedia information. We evaluate our measure on a corpus of multi-structured documents from the INEX 2007 corpus.