Discrete symmetric image registration

Image registration is in principle a symmetric problem. Nonetheless, most intensity-based non-rigid algorithms are asymmetric. In this paper, we propose a novel symmetric deformable registration algorithm formulated in a Markov Random Fields framework where both images are let to deform towards a common domain that lies halfway between two image domains. A grid-based deformation model is employed and the latent variables correspond to the displacements of the grid-nodes towards both image domains. First-order interactions between the unknown variables model standard smoothness priors. Efficient linear programming is consider to recover the optimal solution. The discrete nature of our algorithm allows the handling of both mono- and multi-modal registration problems. Promising experimental results demonstrate the potentials of our approach.