2DPCA Based Nonlocal Means Filter

While The nonlocal(NL)-means filter introduced by Buades et al. have speedily entered the hall of fame of denoising methods, its computational complexity is quite a burden. we propose a newly automatical and efficient preselection-based NL-means filter, using the 2DPCA technique. Mainly, our contributions to the NL-means filter are: 1) a preselection of the most relevant pixels using the 2DPCA technique; 2) adaptive determination of the similar sets. The new approach has been tested on a commonly-used image database and a real noisy image. The results demonstrate that our method significantly improves the denoising effect and the restored have higher peak signal-to-noise ratio.