Investigation of Effect of Noise Parameter on Brain MR Image Restoration

The effect of noise parameter, standard deviation, on Brain MR Image restoration was investigated in order to select the filtering mechanism to get exact restoration. It is a well known fact that the noise in Brain MR image is Rician distributed, but if the standard deviation is between 4.51 to 4.28, the noise is neither Rician nor Gaussian; however when standard deviation is below 4.28 the noise behaves as Gaussian distributed. For accurate restoration, knowledge of this threshold value of σ, where noise characteristics changes, is essential so that corresponding filter can be chosen to remove the noise present in the image. It was found that the characteristics of noise distribution are differentiated on the basis of filtering competency of linear and non-linear filters. Gaussian filter is efficient at low standard deviation, σ <; 4.28 whereas standard deviation σ ≥4.28 Hybrid filter works well; this is verified using Structural Similarity Index measurement (SSIM),Mean Square Error (MSE) and Peak Signal to Noise Ration (PSNR). Hybrid filter is nothing but a series combination of Gaussian and median filter. The performance of Hybrid filter was also compared with Homomorphic filter, which works best for Rician distributed images, and thus verified that Hybrid filter works best for all three types of distribution over given range of standard deviation.. The proposed algorithm was implemented on synthetic images from BrainWeb data base using MATLAB 2015a.