Point correspondence in unstructured nonrigid motion

This paper describes a method for the estimation of point correspondence in unstructured nonrigid motion. Feature points are derived from images of the same scene undergoing nonrigid motion and correspondences are established. The only assumption we make about the motion is that, apart from, global translation and rotation, the local motion of each point is restricted to a circle of radius δ. This is in contrast to existing techniques which assume either rigid motion, non-rigid motion of a known kind (conformal, etc.) or motion with very small local deformations. The point matching problem is viewed in terms of weighted bipartite graph matching. In order to account for the possibility that the feature selector can be imprecise, we incorporate a greedy matching strategy with the weighted graph matching algorithm. The technique is robust and is applicable to many types of problems