Non-rigid image registration through efficient discrete optimization

Discrete optimization offers attractive capabilities for non-rigid medical image registration. Based on previous work on stereo vision with large displacement spaces, we propose a novel non-rigid registration method. The main idea to minimize the computational burden to constant time complexity is to reduce the search space hierarchically from coarse to fine levels. The large search space helps us to avoid local minima in the optimization scheme. We apply our method to a number of challenging registration tasks. Our approach compares very favourably against state-of-the-art registration methods.