Grid refinement in adaptive non-rigid registration

Non-rigid mutual information (MI) registration algorithms with many degrees of freedom (DOF) are quite useful, but they come at high computational cost and have convergence issues. As a remedy adaptive non-rigid registration algorithms, where DOF is increased adaptively (i.e. the grid is refined adaptively), have been proposed. There are at least two ways to refine a grid adaptively: one based on changes in the global measure, the other based on a local measure. We compare these two and show that a local measure method can have better sensitivity to deformations than the global measure. The local measure employed is a novel method using local entropies and local MI.