Graph Centrality on the Registration of Point-sets (Introduction Talk)

 The problem of point-set registration often arises in Computer Vision whenever one needs to match information available in images, such as feature locations, landmarks, or points representing a surface of an object. It is a challenging task in stereo vision, image alignment, medical imaging, and so forth. Many of those problems have also been addressed using graph theory by taking advantage of the structural information presented in graphs. In this paper, centralities are explored in the point-set registration problem. We propose a variant of the Coherent Point Drift (CPD) by integrating the degree, betweenness, closeness, eigenvector, and pagerank centralities in the computation of probabilities during the E-Step of the Expectation Maximization (EM) framework. We analyze the performance of 2D and 3D point-sets under rigid and non-rigid transformations and our results indicate that the alignment error between point-sets has a faster decay when the centralities are integrated when compared with the original CPD algorithm.