Improved image denoising with adaptive nonlocal means (ANL-means) algorithm

A largest matching area approach to image denoising

“…Extensions to the NLM algorithm proposed adaptive spatial support for superior results, by classifying patches as textured or smooth by edge-detection and morphological operations [17], or by clustering the SVD (singular-value decomposition) of the blocks' gradient fields [18], allowing spatial adaptation for each block. An adaptive neighbourhood is selected for each pixel in [19,20], by iteratively increasing the search-window and carrying out NLM denoising, until the L2-risk with respect to the search-window size is minimizedminimizing the variance subject to a fidelity constraint.…”

Locally adaptive bilateral clustering for universal image denoising

“…Following a similar notation as in Ref. 3, the discrete version of the bilateral kernel can be expressed as…”

“…In this experiment, the set of 100 test images are compressed using the JPEG compression routine with the quality parameter Q varies from 5 to 15 in a 5−unit step size 3 . The restoration results are presented as bar plots of average PSNR, MAE, and runtime in Fig.…”

“…Remarkably, the average runtime consumed by Comparison of average PSNR, MAE, and runtime for deblocking compression artifacts using BF [5], LABCF [16], SBF [6], KSRF [15], TF [14], and proposed filter. 3 The quality parameter Q ranges from 0−100, and higher values mean higher quality (e.g., less image degradation due to compression), but the resulting file is larger. [5], (d) LABCF [16], (e) SBF [6], (f) KSRF [15], (g) TF [14], and (h) the proposed filter.…”

“…Over the years, many nonlinear methods have been pro− posed for additive and impulsive noise removal [3,4]. Spe− cifically for the removal of additive noise, the bilateral filter (BF) [5] and its variants [2,[6][7][8], which enforce both spatial closeness and gray value similarity in their denoising, have proven to be a powerful tool for edge−preserving smoothing.…”