I am trying to learn more about processing high noise images in Darktable. This is about the best I can do with this image after a day or so of trying. I would appreciate seeing other approaches.
2017-03-15_07-07-44_2.NEF (25.8 MB)
2017-03-15_07-07-44_2.NEF.xmp (18.3 KB)
Ultimately thereās always a trade-off between retaining detail and getting rid of noise. IMO a bit of āgrainā is fine. Hereās my attempt with darktable 4.8.1ā¦
A single instance of wavelets denoise with almost default settings is in my opinion the best, as it removes all chroma noise and luma noise is pretty good, since it doesnāt over-smooth like default non-local means algorithm.
In this case Iād also suggest slower shutter speed, maybe down to 1/500, as there is not much fast motion to require 1/2500s, but it depends on how good is your image stabilization (600mm is VERY long). That would bring ISO down to roughly 1250.
Here is my attempt. One of the tricks I do with noisy images is to use the details threshold slider with positive values to limit sharpening to details to avoid sharpening noise. I also use the same slider with negative values to protect details from softening effect when applying denoising techniques except for the initial denoised profile module. The preserve shadows slider in denoise profiled can help a lot with noise.
2017-03-15_07-07-44_2.NEF.xmp (15.5 KB)
Thanks for that tip - I like what you have done and will be examining the xmp file.
As the others said, itās always a trade-off between noise and details ā and, I would add, brightness: less brightening, less noise.
You can get very smooth and yet sharp results, but they look fake.
This is your image, slightly denoised in darktable, exported with bounds of 1024x1024:
And then upscaled 2x in chaiNNer, using the RealESRGAN_x2plus model:
Exporting from darktable with bounds of 2048x2048, applying a chain with the 1x_NoiseToner-Poisson-Soft + 1x-Refocus-Cleanly (for sharpening):
You could then maybe merge them (in darktable as an overlay, or in Gimp as layers), brushing in the output from chaiNNer over the original.
Interesting, never heard of chaiNNer. I donāt like the AI look in images, but it may be a useful tool to have in the box.
That also looks very good. I am going to look into chaiNNer. Thanks for the time you took to show what it is capable of.
Aside
Is there an complete idiotās guide to getting started with chaiNNer? My brief foray, a while back, was not successful.
Upscayl is simpler. Donāt know if it is open-source as well as fee. Recently Iāve been using Neat Image, which is not open (and not free, although the demo edition does quite a lot) but is available for Linux.
/aside
Lots. https://www.youtube.com/results?search_query=chainner+upscale
Please open a new topic if needed.
Maybe Iām a little strange regarding noise in pictures. But I prefer to have some grain in, as long as it is fine and regularly. So here is me attempt. Not smooth like a babyās bottom, but a sharp and natural look. Unfortunately sometimes I think I am alone with this opinion.
If itās being looked at full screen on a phone or PC then it likely looks even better this wayš. Not many images are viewed at 100 percentā¦ and not that many are printed so thinking about the final intended viewing conditions might dictate where you place your threshold for grainā¦
Thatās not strange at all. I usually only do chroma denoising, as I find the colour blotches ugly, but luma grain is fine (sometimes I attenuate it a bit, but I prefer it over waxy looking, then over-sharpened images).
Thanks for sharing; great family photo!
Used one instance of denoise (profiled/ wavelets), removing as much chroma noise as possible while softening fine luma noise just a bit.
I have a high tolerance to noise since I come from the days of film. Film never produced results as smooth as a babyās bum so I donāt mind a little noise if it is required to capture the ambience of the low lighting of a scene. I shot my Canon R7 at 32000 ISO often.
Like others, I donāt mind a bit of luma noise. Especially as thereās always a trade-off between reducing noise and retaining fine detail, especially at high ISO (where you also have to deal with a lower dynamic range).
That āfamousā waxy look is the result of smoothing the fine detailā¦
Also, noise becomes a lot less visible when you reduce the output size. As most of my stuff is for screen display only, that in itself āsolvesā a lot of noise issuesā¦
But i do wonder why you used ISO 6400 with a shutter time of 1/2500 sā¦
Not saying that isāt wrong, itās just not a combination Iād have chosen given a choice.
I try to limit ISO to 1600, perhaps 3200, precisely due to noise issues. Often I use auto-iSO limited to the range of 100-800 (or 100-1600 in more difficult conditions).
