MR Image Restoration using Non-Linear Filtering

A method for noise estimation of a Rician distributed MR image at image level is implemented. It is observed that the method gives optimum estimates at low noise (0.15 ≤ a ≤ 0.2). For image restoration purposes, Wiener and Median filters were implemented. Median filter is known to be more efficient in filtering out Rician noise. However, upon observation of residual image, it was found that Weiner filter is competent in edge preservation. On the other hand, it is also observed that (SSIM) of low noise image is more converged towards unity than that of the filtered image. The proposed algorithm was implemented on synthetic images from BrainWeb data base using MATLAB 2015a. Additionally, the study of low noise estimation is extended for higher even order moments and it was noticed that `Q' in noise estimation deviates from unity as order of moment is increased.