The Research Of Medical Images Registration Based On Feature Points And Mutual Information

Medical image registration is a hot topic in image processing, and is the basic issue of medieal image fusion.The research of automatic regisrraction technical and theoretical of medical images is important in the basic medical research, clinical diagnosis, treatment, and efficacy evaluation.In this paper, theoretical and experimental research of key technology and algorithm theory of automatic registration has been made which based on a set of medical image data.Main tasks are as follows:For the characteristics of current medical images, this thesis study the medical image preprocessing in detail.We summarize the methods of image preprocessing which were most commonly used, and verified the effect with medical images.It lay a solid foundation for the image registration.This study focused on medical image registration based on harris feature points, and medical image registration which based on mutual information.The method based on feature points had been study begin with the operator which commonly used in feature point extraction, and then comparative them from accuracy and stability.Then we made some improvements on algorithms through theoretical and experimental research.We selection the gray of neighborhood between feature point as the measure of correlation similarity, and using the affine transformation as the search space.The affine transformation invariance of Mahalanobis distance is used to calibrate the matching of feature point. Finally, it is the parameter refinement.The method which based on mutual information advantages in no need to feature extraction, requires less man–machine interaction, and is easy to realize automatic registration. In the beginning, we mainly study the relative concepts and principles. On this basis, we discuss the interpolation and optimization in detail. Bilinear interpolation and Parital volume interpolation, and Powell optimization and PSO algorithm has been madeed a detailed comparison. In the last, it is the evaluation of the results. Through the analysis of assessment menthods and indicators of evaluation, we analyse the previous two methods from the precision, speed, and noise immunity. The results show that: both methods have characteristics of high registration accuracy, speed and robustness. However, the results will be varying in the different application object.