Non-rigid Medical Image Registration Based on Improved Optical Flow Method and Scale-invariant Feature Transform

A novel non-rigid image registration algorithm is proposed based on an improved version of the traditional variational optical flow model and the extraction of the Scale-Invariant Feature Transform (SIFT) feature points. In this model, the issue of processing the regions of localized disease abnormalities and un-uniform brightness is tackled by using a data term combining the brightness conservation and gradient conservation assumptions. To solve the issue of severe image blurring and the loss of important details caused by the over-smoothing of the traditional optical flow model, an adaptive anisotropic regularization term is used. By extracting the SIFT feature points and using a multi-resolution layered refining, internal fixed-point iteration and coarse-to-fine warping strategy, the issue of registration of medical images with relatively larger deformation and also that of the details registration of medical images which can not be processed by the traditional optical flow method are well resolved. Extensive experimental results show the effectiveness of the model for non-rigid medical image registration.