NON-RIGID POINT SET REGISTRATION USING MULTI-FEATURE AND GAUSSIAN MIXTURE MODEL

We propose a non-rigid point set registration method which consists of a global and local multi-feature based correspondence estimator with a GMM based transformation updating. We first introduce two distance features for measuring global and local structural diversities among two point sets, respectively. With these two features, a multi-feature based cost matrix is then formed to provide a flexible approach to estimate correspondence by minimizing the global or local structural diversities. A GMM based energy function is finally designed for refining the transformation updating, and minimized by the L2 distance minimization. We test the performance of the proposed method in contour registration and real images, and compared against four state-of-the-art methods where our method demonstrated the best alignments in most scenarios.