MRI de-noising using improved unbiased NLM filter

The magnetic resonance images focus on soft tissues, and it is often necessary for healthcare professionals to reach the final conclusion in clinical diagnosis. However, these images are often affected by random noise, which decreases the visual quality and reliability of the images. This paper presents an improved unbiased non-local mean (NLM) filter to solve the de-noising issue in the MRI images. Local statistics of the noise is combined with the NLM filter to design an unbiased NLM filter. First of all, the Gaussian noise information is extracted from the noisy image by performing the wavelet decomposition, statistically modeling the diagonal sub-band wavelet coefficients, and estimating the noise variance by applying the median absolute deviation (MAD) estimator. Next, the Rician noise is removed by applying a NLM filter which averages the noisy pixels by a Gaussian weight factor. Finally, the NLM filtered output pixels are unbiased by applying the noise bias subtraction method for recovering the original pixel values. Our experiments on real MRI and synthetic images demonstrate that promising results that can be obtained much superior than results estimated using existing non-local mean filtering schemes.