Study On Magnetic Resonance Image Average Based On Non-local Means

For MR image sequence, we can get higher SNR from simple averaging the images. But edges will be blurred since organs shift occurred when we scanning in different time. Traditionally, we can do image registration before simple averaging. Because of the process of image registration is complicated and have smoothing effect, the results are not satisfied.In 2005, after researching many denoising method, Buades proposed a new algorithm based on image self-similarity called Non-Local Means. It works well and people are attracted in working on it. In this paper, we plan to propose a weighted average method based on Non-Local Means, we take advantage of structure similarity to find the offset between different images, and correcting it. After that, we do non-local means average to these moving shift corrected images. Experimental results show the algorithm can improve PSNR while having shaper edges.