Estimation based non-local approach for pre-processing of MRI

Pixel intensities of Magnetic Resonance Images (MRI) are known to get corrupted by noise during the acquisition process, resulting in degradation of diagnostic features. Improper estimation of this noise level acts as a performance constraint for noise suppression and subsequent image processing tasks. This paper utilizes a patch based approach where the decomposed patches of MRI with low texture strength are selected on the basis of gradient covariance matrix. These low rank patches are then used to estimate the noise level through Principal Component Analysis (PCA). The estimated noise level has been incorporated in the Non Local Means (NLM) approach to improve the noise suppression. Simulation trials have been conducted for different noise levels and performance evaluation has been carried out using suitable evaluation metrics. Further, a performance comparison is also carried out for the NLM approach with and without noise estimation; to demonstrate the significance of estimation prior to non-local pre-processing.