Re: [darktable-dev] Another denoising algorithm, looking pretty interesting

2017-12-05 Thread François Tissandier
Interesting. The BM3D method also looks quite good ! Sometimes better than their new method, and always better than non local means. François On Sat, Dec 2, 2017 at 12:33 PM, Sarge Borsch wrote: > https://www.cs.tut.fi/~foi/GCF-BM3D/BM3D_TIP_2007.pdf > And according to test images presented

[darktable-dev] Another denoising algorithm, looking pretty interesting

2017-12-02 Thread Sarge Borsch
https://www.cs.tut.fi/~foi/GCF-BM3D/BM3D_TIP_2007.pdf And according to test images presented in another page (https://dmitryulyanov.github.io/deep_image_prior), "BM3D" might be even better than Non-Local Means. Although it's unclear which patch size for NLM was used in that comparison… if it was