摘要

In this paper the deformation invariant curve matching problem is addressed. The proposed approach exploits an image pyramid to constrain correspondence search at a finer level with those at a coarser level. In comparison to previous methods, this approach conveys much richer information: curve topology, affine geometry and local intensity are combined together to seek correspondences. In experiments, the method is tested in two applications, contour matching and shape recognition, and the results show that the approach is effective under perspective and articulated deformations.