3D object retrieval based on visual keywords using Relative Angle Context Distribution

In this paper, we proposed a new 3D object retrieval method based on the visual keywords. In our method, the visual keywords are generated from the clusters of Relative Angle Context Distribution, which provides a statistical shape context that captures local shape characters and is also rotational and scale invariant. We also adopt the term frequency model commonly used for text retrieval to compare two 3D objects according to their keyword sets based on a cosine similarity measure. Our experiments have demonstrated that this method is robust for its invariance with respect to object orientation and scaling, and yields a higher precision when compared to existing retrieval methods such as the Distance Map.