Fast Non-Local Mean for Image Denoising

This paper presents an efficient image denoising method by using Gaussian Principle Component Analysis (GPCA) in conjunction with Non-Local Means (NLM). By taking into account of the noise feature, this method pre-filters the image before PCA analysis to improve the PCA efficiency and decrease the dimensionality of projected the vector. By applying GPCA to project the image patch into a lower-dimensional (LD) space, the accuracy of the similarity weights calculation for NLM can be improved with less computational complexity. Experiments show that our method performs well in terms of image visual fidelity as well as PSNR with a degree of detail preservation.