Abstract
Proliferation of 3D shape models has prompted study into shape similarity search retrieval of 3D models. Descriptors for a 3D shape model need to be compact in order to reduce its storage cost and computational cost for shape similarity comparison. On the other hand, a shape descriptor having better retrieval performance tends to have a large dimension. In this paper, we explore an approach to reduce storage and similarity comparison costs while maintaining retrieval performance. Specifically, we evaluated the effects of principal component analysis (PCA) and its non-linear variant, kernel-PCA, for their effectiveness in reducing dimensions of shape descriptors without sacrificing retrieval performance.
