Abstract:

After representing the deformation as a field of discrete displacements, the nonrigid image registration problem can be considered as a maximum a posteriori inference problem in Markov random field, which can be optimized using pairwise Gibbs energy minimization technique. To find a minimum, the registration problem is modelled as a game among pixels, a novel approach for nonrigid image registration based on cooperative game theory is proposed, and a deterministic algorithm is designed. During the game, each pixel tries to form a coalition with its neighbors, and select a displacement to maximize the payoffs of itself and the coalition. Registration is achieved by an iterative process. The novel approach can register complicated deformation with high accuracy and good robustness, so it is very suitable for fine registration. Experimental results show that the cooperative game theoretic approach is feasible and effective.