Help with wavelets mode on denoise (profile)

Hi, does anyone have any tips for using the wavelets mode of the denoise profile module on darktable? I use non local means most of the time, but I prefer wavelets for portraits to preserve more skin details.

I have watched videos on it and I think I got how it works in general, but I usually spend too much time fiddling with the curve until I find something that suits the picture. I would like to know if anyone has a good preset that could be used as a starting point instead of the standard “flat line”. Any tips on how to make a simpler/better use of this tool would be welcome!

2 Likes

There’s a bit of a writeup in this draft article, which is based on the info from @rawfiner’s videos, maybe it is useful:
https://elstoc.github.io/dtdocs/module-reference/processing-modules/denoise-profiled/

1 Like

I’ve created a number of presets in Y0. The presets apply different levels of strength 1, 2, 3, and pull down the fine details by different amounts 1, 2 or 3 lines, per strength. That way I can just toggle through the presets for each image. However I learnt something reading the document @Matt_Maguire linked too - the R,G,B method looks interesting if one has the time.

The R, G, B method looks like it could benefit from a visualisation tool so that you can view the channels while setting their curves, without having to go to the channel mixer module and back.

Bruce Williams explained/showed this channel mixer technique way back when in episode 45 (from 07:17 onward).

This is about the raw denoise module in 2.6, but the technique is still valid when using RGB wavelet denoising in denoise (profiled)

1 Like

Ah, didn’t catch this one. But it was Bruce’s video on Denoise Profiled that first helped me understand the wavelets tool (coupled with the important top comment by @rawfiner

1 Like

Rawfiner video on 2.6 does and extensive review of denoising by channels esp with the bilateral filter…

Thanks for all the references, I already learned a lot!

Just another question. When denoising portraits, it always seems to me that wavelets also preserves skin colors better than non-local means. Is there anything I could do to obtain similar results when using non-local means?