Researches on 3D model automatic annotation and retrieval based on semantic correlation

Most of current researches about 3D model retrieval, one important part of multimedia information retrieval, are mainly in the shape-based way. However, their effects are not satisfactory. This paper proposes a semantic-based 3D model retrieval method which uses the semantic correlations among 3D models to assist automatic annotation and semantic classification. Firstly we measure each model´s semantic property by its semantic relation with the others. Then we use clustering technique to find out these models having strong semantic correlations and name them for semantic community. Finally based on the semantic community, we automatically annotate 3D models´ semantics. The experiments on Princeton Shape Benchmark datasets show that our method achieves good effect on all the following aspects: semantic annotation, semantic classification and retrieval.