A matching algorithm based on hierarchical primitive structure

A novel motion matching algorithm based on a hierarchical primitive structure (HPS) is presented. It has two main steps. The first is the construction of the HPS. Zero crossings are extracted using a Laplacian or Gaussian operator, and they are assembled according to certain rules. The second is a coarse-to-fine matching step based on similarity criteria. A strategy which combines relaxation and constrained tree-searching is used to obtain the optimal matching result. Experimental results for real images are given.<>