MRI (Magnetic Resonance Imaging) images suffer from noise due to long acquisition time and high resolution is needed for post processing the noisy image for further analysis like segmentation and registration, which is required for the clinical practice for better analysis of disease detection. In MRI, the noise from the acquisition device follows the complex noise distribution. For better analysis of MRI data, the complex noisy data follows the magnitude data which obeys Rician Distribution. So the magnitude data need to be denoised rather than the complex data in which, it suffers from phase artifacts. Unfortunately, the presence of Rician noise in these images affect edges and fine details, which limit the contrast resolution and make diagnostic more difficult. The key idea from using diffusion tensor is to adapt the flow diffusion towards the local orientation by applying anisotropic diffusion along the coherent structure direction of interesting features in the image, which enhances the quality and improves the Signal-to-Noise Ratio (SNR) and edge preservation of anatomical details. In this paper, the filtering parameter is automatically chosen from the estimated standard deviation of noise using standard Linear Minimum Mean Square Error (LMMSE) method in which, it improves the convergence rate of diffusion along with preservation of anatomical details which is required for clinical diagnosis. Also, matrix extension of the scalar diffusion filter, is proposed which automatically adapts to the local structure of the image and the level of noise along with the coherent structure direction of interesting features in the image, which improves the SNR and edge preservation of anatomical details.

In order to illustrate the effective performance of the algorithm, some experimental results are presented on synthetic and clinical images. The proposed filter shows better preservation of edges and efficient noise reduction at both low SNR and high SNR levels.