Non Local Image Denoising Using Image Adapted Neighborhoods
In recent years several non-local image denoising methods were proposed in the literature. These methods compute the denoised image as a weighted average of pixels across the whole image (in practice across a large area around the pixel to be denoised). The algorithm non-local means (NLM) proposed by Buades, Morel and Coll showed excellent denoising capabilities. In this case the weight between pixels is based on the similarity between square neighborhoods around them. NLM was a clear breakthrough when it was proposed but then was outperformed by algorithms such as BM3D. The improvements of these algorithms are very clear with respect to NLM but the reasons for such differences are not completely understood. One of the differences between both algorithms is that they use adaptive regions to compute the denoised image. In this article we will study the performance of NLM while using image adapted neighborhoods.
- 2.Boulanger, J., Kervrann, C., Bouthemy, P.: Adaptive space-time patch-based method for image sequence restoration. In: Workshop on Statistical Methods in Multi-Image and Video Processing, SMVP 2006 (May 2006)Google Scholar
- 4.Buades, A., Coll, B., Morel, J.M.: Denoising image sequences does not require motion estimation. In: Proc. IEEE Conf. on Advanced Video and Signal Based Surveillance, pp. 70–74 (2005)Google Scholar
- 8.Olsen, S.I.: Noise variance estimation in images. In: Proc. 8th SCIA, pp. 25–28 (1993)Google Scholar