An adaptive non-rigid image registration technique using hierarchical B-splines

An efficient non-rigid image registration framework aims at accurately constructing a spatial transformation, which enables the smooth deformation of an image in order to correctly align it with the target image. We present a new approach to implement non-rigid image registration, in which the spatial transformation is constructed by using hierarchical B-spline basis functions. Based on the concept of minimization of an energy functional, which drives the registration process by minimizing the difference between the images, the control points are dynamically updated resulting in a new spatial transformation function at each stage of the registration process. Moreover the registration is carried out on a set of successively refined grids, which makes the method more efficient and faster. Instead of uniformly refining the entire control grid, this dynamic implementation is carried out by an adaptive grid refinement procedure, wherein we detect large deformation areas and locally refine these regions. This makes the computation of the spatial transformation more efficient and accurate with fewer number of control points. A comparison between the adaptive grid refinement and uniform grid refinement methods is explained using synthetic and medical images as examples.