A quadratic programming based cluster correspondence projection algorithm for fast point matching

Point matching is a challenging problem in the fields of computer vision, pattern recognition and medical image analysis, and correspondence estimation is the key step in point matching. This paper presents a quadratic programming based cluster correspondence projection (QPCCP) algorithm, where the optimal correspondences are searched via gradient descent and the constraints on the correspondence are satisfied by projection onto appropriate convex set. In the iterative projection process of the proposed algorithm, the quadratic programming technique, instead of the traditional POCS based scheme, is employed to improve the accuracy. To further reduce the computational cost, a point clustering technique is introduced and the projection is conducted on the point clusters instead of the original points. Compared with the well-known robust point matching (RPM) algorithm, no explicit annealing process is required in the proposed QPCCP scheme. Comprehensive experiments are performed to verify the effectiveness and efficiency of the QPCCP algorithm in comparison with existing representative and state-of-the-art schemes. The results show that it can achieve good matching accuracy while reducing greatly the computational complexity.