Robust Non-rigid Point Set Registration Based on Global and Local Mixture Structural Feature

This paper proposes a mixture structural feature based non-rigid point set registration algorithm, this algorithm mainly contains correspondence estimation and space transformation. In the step of correspondence estimation, Euclidean distance and angle based global and local structural features are used to describe structural difference between two point sets. Then the global and local structural features are combined to construct a cost function, which provides the correspondence by minimizing structural difference using a linear assignment method. In the step of space transformation, a thin plate spline function is used to solves space transformation problem of the source point set. The performances of the proposed algorithm are verified by simulation data and real data, and compare with four classical algorithms where proposed algorithm achieves the best alignment in most cases.