In this paper, a medical image registration method using points, contour and curves is proposed, which has the accuracy of registration based-on points and the robust of registration based-on lines (including contour and curves). The operator can exactly extract the features from the images with semi-automatic extraction method. The extracted curve-pairs are modeled by non-uniform cubic B-splines so that they have the same parameter space after modeled. Then the mechanism of non-uniform subdivision of curves ensures that the subdivided points can match the original curve as close as possible and fulfill the requirement of image registration. The deformation between images is described by thin plate splines, which is solved iteratively by continuously improving the regions of maximum difference on contour-pair or curve-pairs and adding constraint conditions. The experimental results show that our algorithm, which exploits the advantages of registration based-on points and that based-on lines, is a robust, fast and accurate medical image registration algorithm.