3D model retrieval using multiple features and manifold ranking

The demand of 3D object retrieval became urgent according to the widely use of 3D printer. In this paper, a 3D object retrieval method is proposed using multiple features and manifold ranking. Five descriptors are concatenated to be a new feature vector of length 792 for 3D object retrieval. They are angular radial transform-based elevation descriptor(ART-ED), principal plane descriptor(PPD), 3D-angular radial transform(3D-ART), shell grid descriptor(SGD), and Grid Distance 2 (GD2). Next, a manifold ranking method is used to re-rank the retrieved results. In addition, various distance metrics are addressed in the construction of manifold graph. The retrieval results on a benchmark dataset SHREC-W have been reported to show the feasibility of the proposed method.