Nonlocal variational method application for image denoising

Combining of nonlocal means and variational models can extend the ability of image processing such as texture image processing. In this paper, an implementation scheme of nonlocal variational models is proposed, which contains discretization, coordinate transformation, and dimension expansion. Firstly, continuous models are transferred to discrete model. Then we transform absolute coordinate to relative coordinate. Because relative coordinate can limit to patch area instead of whole image area, it enormously reduces the data storage. Finally we expand dimension of some array, so it is easy to calculate directly in MATLAB because arrays become same dimension. In experiment, we add noises such as Possion, Gaussian, Salt&Pepper, and Speckle to original image. The results show the proposed scheme is validated for all kind noises and is quite efficient.