Graph Correspondence-Based Point Set Registration

Liang Li1, Ming Yang1, Chunxiang Wang2, Bing Wang1

  • 1Shanghai Jiao Tong University
  • 2Shanghai Jiaotong University

Details

10:30 - 13:00 | Tue 22 May | LBP Zone | TuA-LBP.1

Session: Late breaking posters - 1

Abstract

Correspondence is a major problem in point set registration and it also leaves an open problem in constructing correspondence between point sets. In this paper, the point sets are denoted by graph, then the correspondence problem between point sets is converted to graph isomorphism problem. This problem can be solved by effective probabilistic linear programming heuristics. After establishing correspondence between point sets, the point set registration is equal to absolute orientation problem, which is a classical problem in computer vision or pattern recognition and can be solved by singular value decomposition (SVD). This registration algorithm proceeds in an iterative optimization manner.