Point Set Registration of Large Deformation Using Auxiliary Landmarks

Abstract

We present a probabilistic non-rigid point set registration method to deal with large and uneven deformations. The registration is treated as a density estimation problem. The main ideas of our method are to add constraints to enforce landmark correspondences and preserving local neighborhood structure. Landmarks represent the salient points in point sets, which can be computed using feature descriptors such as scale-invariant feature transform. By enforcing landmark correspondences, we preserve the overall global shape of the point set with significant deformations. In addition, we incorporate constraints to preserve local neighborhood structure by leveraging Stochastic Neighbor Embedding (SNE), which penalizes incoherent transformation within a neighborhood. We evaluate our method with both 2D and 3D datasets and show that our method outperforms the state-of-the-art methods in a large degree of deformations. In particular, quantitative results show our method is 49% better than the second best result (from the state-of-the-art methods). Finally, we demonstrate the importance of using correct landmark correspondences in registration by showing good registration results in large and uneven deformations point sets.