3D non-rigid pattern recognition based on structural analysis

This paper presents a new retrieval approach for 3D non-rigid objects. The proposed method consists in using the Reeb graph representation as local shape descriptor. The generated Reeb graph, based on the heat diffusion properties, is segmented into Reeb charts having a controlled topology. Each Reeb chart is associated with a couple of geometrical signatures, based on the area and angle distortions. The matching procedure is carried out on each pair of Reeb charts, according to the minimum distance between the corresponding signatures. A global similarity measure quantifies the degree of correspondence between all the matched Reeb charts. Experimental results, conducted on SHREC 2011 dataset, have shown that our retrieval approach provides an overall retrieval efficiency gain compared to three state-of-the-art methods.