Primal-dual optimization of the fractional-order variational optical flow model for image registration

Image registration is one of the fundamental and essential tasks for image fusion and image analysis. However, now there are still some problem exiting in non-rigid image registration. For example, in large displacement condition, the image registration accuracy is not satisfied. To deal with it, we present a novel fractional-order variational method to estimate optical flow, using the primal-dual method to construct convex function for global optimization, and applying pyramid iteration to calculate large displacement. Numerical experiments show that the proposed approach can not only obtains more accuracy than classical optical flow method but also works well in large displacement condition. It also demonstrates that fractional-order method as generalization of integer-order method has substantial improvements in image registration accuracy.