Robust Point Matching Using Mixture of Asymmetric Gaussians for Nonrigid Transformation

In this paper, we present a novel robust method for point matching under noise, deformation, occlusion and outliers. We introduce a new probability model to represent point sets, namely asymmetric Gaussian (AG), which can capture spatially asymmetric distributions. Firstly, we use a mixture of AGs to represent the point set. Secondly, we use L2-minimizing estimate (L2E), a robust estimator to estimate densities between two point sets, to estimate the transformation function in reproducing kernel Hilbert space (RKHS) with regularization theory. Thirdly, we use low-rank kernel matrix approximation to reduce the computational complexity. Experimental results show that our method outperforms the comparative state-of-the-art methods on most scenarios, and it is quite robust to noise, deformation, occlusion and outliers.