We propose a novel superpixel-guided nonlocal means (SNLM) method.

The performance of SNLM is better than the existing NLM method.

SNLM adaptively selects the k neighbors in a flexible search region and does not require an empirical study on the search window size.

SNLM is effective to improve the performance of the classic image denoising and image super-resolution algorithms.

The thorough quantitative and qualitative results demonstrate that SNLM achieves better results than state-of-the-art algorithms.