OPTIMAL DISCRETE MULTI-RESOLUTION DEFORMABLE IMAGE REGISTRATION

Deformable image registration is a challenging problem. Current solutions either have a high computational load or suffer from weak regularization. We propose a graph-based deformable registration technique that exploits dynamic programming to optimally solve a discrete multi-resolution formulation. It allows the use of regular similarity metrics known in literature and incorporation of local deformation models. We explain the theoretical framework behind the optimization and show results of registrations on synthetically deformed 2D brain CT images, with and without the multi-resolution deformation model. The proposed method shows the effectiveness of the optimal multi-resolution optimization and the promises for future extension to 3D.