View topics: Automatically generated characteristic view for content-based 3D object retrieval

Characteristic view is an effective way to represent a 3D object through a set of distinct projections from different view aspects. In this paper, we proposed techniques for automatic characteristic views generations by clustering views of the object from multiple view aspect. By considering the resulting clusters as View Topics that describe a set of portraits of the object, the object can be represented by a set of view topics that can be applied to 3D object retrieval with similarity measures based on the Vector Space Model and the Language Model as well as advanced techniques such as RBF Kernel method. Our experiments have demonstrated that our method is not only invariant with respect to rotation and scaling, but also invariant with respect to the object reflection, and achieve an overall better performance than existing methods. Copyright © 2009 ACM.