The Application Of Rotation Invariant Non-local Means In Magnetic Resonance Image Denoising

In clinical Magnetic Resonance Imaging(MRI)applications,we often average from multiple scans to increase the signal-to-noise ratio of magnetic resonance images.However,averaging the images will often blur the edge and details of organs in due to the unavoidable movement of body.To solve this problem,our group proposed one method base on non-local means(NLM),using the similarity between patches to derive the local offset between the images,and then averaging offset-compensated image patches to get the final image.In this paper,we propose a rotation-invariant non-local means(RINLM)which uses circular patches consisting of a series of rings with equal area instead of square patches to search similar patches in images.Compared with NLM,RINLM can find more similar patches in images which contain many rotated local structure.In MR imaging,there are many local rotated motions between images.So,this method can be used in sequence averaging and denoising to greatly improve the SNR of such images.The experimental results show that proposed method can increase the quality of image.