Study on non-rigid medical image registration based on optical flow model

Abstract

A new method for image registration based on step-by-step registration is proposed. First, SIFT feature algorithm is used to image to get rough registration then optical flow algorithm is used to achieve the final accurate registration. Image feature point extraction is the basis of medical image registration and its accuracy impacts on the matching results directly. SIFT feature algorithm is based on local image features and has a good feature of scale, rotation, illumination invariant. Optical flow algorithm no need to do feature extraction and this method calculation is quick and simple and it uses image intensity information directly. The two steps are complementary that the former prepare for the latter to improve the convergence rate while the latter allows a more accurate registration result. The experimental results show that this algorithm can improve the efficiency of non-rigid medical image registration accuracy and increase convergence speed thus it has certain superiority in terms of image registration.