Multiscale nonlocal means for image denoising

The non-local means method (NLM) is widely used in image denoising. However, the performance of this method heavily depends on the choice of smoothness parameters. In this paper, we present a novel multi-scale non-local means method (MNLM) for image denoising. By introducing the multi-scale decomposition of images, our method can avoid the difficulty of choosing the smoothness parameters. Compared with the classical NLM method, MNLM not only improves the accuracy of the measurement of similarity, but also generates better denoising results.