A Clustering Successive POCS Algorithm for Fast Point Matching

This paper proposes a clustering successive projections onto convex sets (CSPOCS) algorithm for enforcing two way constraints originated from point matching. Via point clustering, the problem of projection onto convex set (POCS) where the convex set is described by point correspondence's constraints is converted to the POCS problem where the convex set is described by cluster correspondence's constraints. Then the successive POCS (SPOCS) technique is employed to solve the POCS problem. The resulting algorithm can be viewed as a generalization of SPOCS by combining with clustering. Its precision and computational load are decided by the average radius of clusters. Experimental results demonstrate the effectiveness of the algorithm