Sparse Representation for Robust 3D Shape Matching

With the number of 3D shapes has risen sharply, a fast and robust matching technology suitable for large 3D shape databases is one of the key technologies to enhance the retrieval performance. We proposed a general novel matching algorithm for 3D shape retrieval: SRRSM, based on sparse representation of signals. Using feature database of 3D shape as over-complete dictionary, the matching problem can be transfer to the problem of sparse representation of signals. It is a second-cone programming (SOCP) problem and can be solved in polynomial time by interior point methods. The proposed approach combines signal reconstruction, sparse and discrimination power in the objective function for matching. It is more sparse and robust for effective matching than the Euclidean distance the most commonly used for matching. Meanwhile, the proposed method is very suitable for large 3D shape database. Theoretical analysis and comparative experiment verify the efficacy of the proposed algorithm.