Nonlocal Patch Functional Minimization for Image Denoising Using Nonsubsampled Contourlet

The nonsubsampled contour let transform (NSCT) is an excellent multi scale and multidirection representation for images, born in redundant and shift-invariant quality suitable for denoising. Most NSCT or NSCT-like denoising methods borrow the mature algorithms from wavelets, and are restricted by the precision of the prior model to describe the coefficients statistics. This paper presents a discrete regularization approach relying on the nonlocal weighted patches function and the NSCT sub band estimator, to relieve the effect of prior precision while suppressing the additive noise and the resultant artifacts. Results on the PSNR and vision comparisons with the advanced denoising algorithms demonstrate the superiority of the proposed method.