MRI denoising using pixel-selection-based non-local means

The non-local means algorithm has been proven to allow a good image restoration performance of MR images. However, this method utilizes all pixels of the images, some of which might be useless for noise removal. In this paper, an improved pixel-selection-based non-local means algorithm which avoids the usage of useless pixels is supplied. The principle of pixel selection is based on the similarity between neighborhoods of the median value and mean absolute error which are insensitive to noise. The denoising performance is evaluated through quantitative methods (peak signal-to-noise ratio and mean structure similarity) and qualitative methods (a visual comparison of image residuals and denoised images). From the results over synthetic and real images we can conclude that the pixels-selection-based non-local means algorithm is of overall satisfaction for MR image denoising.