Denoising high-ISO images: a sourdough bread

So you could do something similar…just do your denoise blend in difference mode to confirm what you are impacting and then turn it off…do your edit and then turn it back on again…or you could even use raw denoise initially and you would not have to be too concerned with tweaking the setting for softness with your approach as you over denoise…then in the end turn off the raw denoise and apply your choice of noise reduction…

yeah that’s mostly how I work now, if the files aren’t preprocessed by DxO or something.
Crank up raw-denoising to get a (way to) smooth of an image, but it helps in setting things like exposure and filmic without getting too distracted on little peaks of data, and it makes the filmic auto-button work better.

Then when it’s set to a good start point I turn off raw-denoising, use a chroma-only denoise setting and leave it at that.

A while ago I was thinking about how to improve the darktable auto min / max color pickers, so they wouldn’t always pick an outlier pixel, mainly because I was having trouble with the negadoctor pickers (More robust color picker in Darktable). Probably the simplest solution would be to have the picker pick the min/max from a blurred version of the preview image, maybe with something like a median filter applied.

I didn’t get any good result with dt. I used Exporting/importing photos in full 32-bit representation (to train a neural network for image denoising) but it is not trained on Olympus raw file, so there where some artifacts. I exported another file from dt and merged them to get some more structure and less visible artifacts.

A long time ago it was recommended to turn on raw denoise esp for the black level autopicker in filmic…then just turn it off…

Since this thread takes a close look at denoise, I’ll bring up an issue I am seeing, but if you think I should make a new thread, I’ll do that.

The issue is that when I export this edit, there is noticeably more noise in the jpeg than in the image shown in dt.

I am using darktable-3.7.0+795~g0cae2c24d-win64.exe.

“That’s the piece I wanted!”

IMG_7275.CR2 (20.8 MB)

IMG_7275_04.CR2.xmp (38.5 KB)

The posted jpeg is what you get with denoise on. If you turn off denoise and export, there is not much difference in noise. I have tested turning all other modules, except exposure, and they don’t change the increase in noise.

I see a pretty massive difference on vs off??



This sounds peculiar || interesting.
What export-to-jpg-settings do you use?

Have fun!
Claes in Lund, Sweden


I’m trying to say that I see more noise between the image I see in dt versus the exported jpeg when I am using denoise than when I am not. There is a lot more noise in the image in dt with denoise turned off then when it is on, but that is not the issue. It is between the image and the exported result.

dt with denoise turned off

jpeg with denoise turned off

dt with denoise turned on

jpeg with denoise turned on

Now, I might not be able to see the difference between the dt image with denoise turned off and the jpeg because there is so much noise to begin with. Therefore, the problem might not be with denoise and just with export.

Only “odd” thing I have detected so far:
why do the four last images you posted
differ so much in file size?

I don’t know. I’m using Windows Print Screen on all four.

After looking more at the actual images, I think it is not a denoise issue, just a jpeg issue. I think I need to take this off this thread, so as not to hijack it.

Can you export tiff and compare noise to darktable?
Then, can you export from that tiff to jpeg in another program, at 100% jpeg quality, and compare that to a jpeg export from dt at 100% quality?

Makes no difference; tiff is just the same as jpeg.

Try 95 or 100 but not 99…Strangely I feel I recall some other issue in the past with JPG artifact if someone tried 99%?? I just can’t recall the exact issue…

1 Like

No difference I can see.

I exported your edit and then in DT I scrolled to 100% and took a snapshot of the raw and then I advance to the jpg and compare …I really don’t see a lot of difference and what I do see is actually very marginally but less noise in the jpg…at least zoomed in and looking at the face of the bald gentleman??? I uploaded by exported jpg from your edit…

I can’t quite grasp your explanation of what you did, but here is a shot of the shirt on the right exported at 100%:


This is from dt:


So I use LittleCMS enabled and I dont use prefer performance to quality…not sure if this impacts in any great way what the preview in DT looks like… Basically i downloaded your canon file and xmp and did a strait export…so this is noise Reduction on…Then I rolled the scroll wheel in place to zoom in to 100% , took a snapshot…then advanced to your jpg export or at least the one done on my pc…it is automatically at the same zoom and location …so now the raw is to the left and the jpg is to the right…sliding the snapshot line across I can see very little difference…how are you viewing your jpg??