Research Of Image Registration Based On Mutual Information And B-spline FFD Model

Image registration is a necessary step before image fusion, it is a geometric transformation by seeking in different imaging devices, shooting angle, shooting the same scene in time to collect two(or multiple) images to pair accurate image processing. Image registration can be divided into rigid registration and non-rigid registration of two categories, rigid registration mainly to solve simple problems overall movement of the image, and algorithms related research is relatively mature; non-rigid registration emphasis on solving complex non-image linear transformation issues related research is still in the development stage. How to effectively deal with the image of a non-rigid transformation issues, to enhance the accuracy of matching algorithm is the key to our study.In this paper, we research the image registration algorithm based on mutual information. This algorithm does not require image segmentation, and other preprocessing operations, the grayscale image information can be achieved only based on registration, its advantages that the implementation process is simple, good robustness. Based on the research of mutual information theory, combining the POWELL improved algorithm is proposed based registration algorithm largest mutual information, in the registration of the simulation brain image, results can reach sub-pixel.In order to solve the problem of nonlinear distortion in the non-rigid image registration exists, we studied the free deformation model based on B-spline. Use of the B-spline has the advantage of good local properties, while sparse grid control can be achieved overall image registration, relatively fine control mesh allows the image to achieve a more precise registration. In this paper, the theory of mutual information and free registration B-spline transformation model, we propose a mutual information with the constraint B-spline free elastic deformation image registration algorithm, first, rough global image registration, get the image registration parameters, and then combined with LBFGS optimization algorithm, constantly adjusting the control mesh to give a non-rigid registration optimal parameters, to enhance registration algorithm accuracy.The proposed method MATLAB 2013 A and Visual Studio 2010 as a platform to verify the effectiveness of the algorithm. A combination of mutual information and normalized mutual information image registration two measures of effectiveness objective analysis, through several experiments show that the proposed method successfully on non-rigid image registration, effectively increasing the registration accuracy and successfully corrects nonlinear distortion present in an image.