The registration of medical images with local deformation is a difficult problem, especially the images to be registered not only have local deformation but also have great difference in gray-scale distribution. It is a excellent method to solve this problem by adding the Euclidean Distance Metric based on landmark points as the penalty term to the Intensity-based metric in the registration. In this paper, an improved distance measure is proposed, which makes the local deformation more accurate than the Euclidean Distance Metric. In addition, for the multi-resolution registration model, an adaptive weight of metric method for multi-resolution registration is proposed, which makes the registration results better.