Non-rigid point set registration via coherent spatial mapping

Point set registration is essential for many computer vision technologies. In this paper, we proposed a new point set registration method based on coherent spatial mapping (CSM), whose solution is formulated as a maximum likelihood problem. Our algorithm iteratively recovers the point correspondence and estimates the transformation between two point sets. In the first step of the iteration, feature descriptors such as shape context are used to establish rough correspondence. In the second step, each training sample is associated with a latent variable which indicates whether it is an inlier, and then the thin-plate spline (TPS) is chosen to parameterize the CSM and Expectation Maximization (EM) algorithm is used to solve it. The registration experiments are undertaken on various synthetic and real data. The results demonstrate that the proposed approach is robust and accurate for scenes under deformation, noise, outlier, occlusion, and rotation and is superior to the state-of-the-art methods.