Medical Image Non-Rigid Registration Method Guided by a Matched Feature Point Pair

When using optimization techniques to solve non-rigid image registration, improper initial values often cause the optimization process to converge to a local minima and lead to a failed image registration. In order to solve this problem, we propose a new non-rigid image registration procedure which is guided by a pair of matched feature points. Firstly, an initial local region is determined according to the pair of matched feature points. Secondly, the local registration region expands gradually to cover the whole image as the optimization process continues. In the process that the initial local registration region is determined and expanded, the values of the deformation parameter vector are estimated according to the spatial locations of the matched feature point pair and the statistical prior that the deformation field is a Gaussian Markov random field (GMRF). The parameter vector of the deformation field in the local registration region is solved by using the registration method based on the image's intensity statistical information. Experimental results show that the proposed approach can effectively overcome the problem that the non-rigid registration method is liable to be trapped in a local minimum.