Non-Rigid Point Set Registration via Gaussians Mixture Model with Local Constraints

Abstract

The local feature of point set is as important as the global feature in the point set registration problem. In this paper, a non-rigid point set registration method based on probability model with local constraints was proposed. Firstly, Gaussian mixture model (GMM) is used to determine the global relationship between two point sets. Secondly, local constraints provided by k nearest neighbor points helps to estimate the transformation better. Thirdly, the transformation of two point sets is calculated in reproducing kernel Hilbert space (RKHS). Finally, expectation maximization (EM) algorithm is used for maximum likelihood estimation of parameters in our method. Comparative experiments on synthesized data show that our algorithm is more robust to distortion, such as deformation, noise and outlier. Our method is also applied to the retinal image registration and obtained very good results.