I’m using Darktable 3.2.1 and the new “scene-referred” workflow with filmic v4. I’ve noticed on several water scenes that I’ve taken with my Canon 50D and Sigma 18-50mm f/2.8 EX DC that there appears to be noise in the sky and water, even at ISO 100. In the water_noise.png example, the noise appears to be red or green pixels whereas the sky_noise.png example appears to be lighter or darker pixels. I can correct both of these using a blur on the ab channels of LAB with the lowpass module (see the _fixed.png attachments), but I have a couple questions:
is there a preferred way to do this using the new filmic workflow (linear RGB)?
why is this noise occurring at such a low ISO? Is this just a result of the limitations of the sensor on the 50D (e.g. would a newer sensor resolve this issue)? Or is this more of a limitation of the APS-C sensor size (and full frame would help)?
That doesn’t appear to make a difference with sky_noise.png but does help with water_noise.png (left is using PPG, right is using AMaZE with color smoothing set to “five times”):
In new filmic v4 ‘add noise in highlights’ is by default set to 0.1. Try turning it to 0 and see if it makes a difference - you can find this in the ‘options’ tab.
I agree, the green and red edges in your detail enlargement point to chromatic abberation. Even if your lens is not supported you can try to use module “lens correction” to correct. Select a lens which is similar to yours, then
First, I think you have quite an old camera whose sensor is noisy.
Second, noise is normal, even with new cameras, according to my experience there is always some noise. If there is no noise, some kind of noise reduction is active.
Third, noise depends on your workflow and how you edit your photos. If you edit your photos, you will amost always create some noise. So noise anlways needs to be reduced.
Yes, it is totally normal that there is noise even in ISO100 shots.
I don’t think that full frame sensor produce less/no noise at all, although I must admit that I never had a full frame camera.
Noise belongs to a digital photo. Having no noise at all is an ideal.
It is totally normal to have some chroma noise in a bright blue sky at ISO100 at good lighting conditions.
this is only active if highlight reconstruction is active - the default threshold setting in reconstruction tab disables this.
So in fact the noise seems to be there in the raw - but thats guessing without having seen the raw file
if it’s just chroma noise, the denoise (profiled) in non local mans mode (or non local means auto) with setting central pixel weight (details) to 1.0 might be an option to suppress the chroma noise
Good idea but unfortunately this didn’t make a difference
I agree that the red and green in the white water breaking on the cliffs definitely does, though the chromatic aberrations module doesn’t seem to help. Using lens correction with my lens and the above settings (and slightly adjusting TCA red and TCA blue) appears to improve the colors on white water:
Such a tool exists here: https://www.photonstophotos.net/
Click on the read noise chart, then select the cameras you want to compare from the list on the right.
Btw, how is the exposure in these noisy photos (before any adjustments)? Did you ETTR, or is most of the data over to the left of the histogram? You may know already that noise is primarily a factor of light, not ISO, so it is possible to get noisy images at 100 ISO if the image is underexposed.
As others have suggested, you can’t really avoid noise completely. I find chroma noise much uglier than luma noise. To eliminate only chroma noise in darktable I use denoise (non local means) module, and turn luma to 0. Not sure if there’s a better method but it works for me.
The CA module has never really helped me, but lens correction and defringe have (check out the settings for defringe - check out the tooltip!). With defringe, sometimes I use a drawn mask to only process the (most) problematic areas, while preserving detail in the rest of the photo. That way, I can crank up the correction strength.
I can’t see any noise in the water. I can see variation, but I expect that. The water has waves at all kinds of angles, and these reflect different portions of the sky. And waves have little wavelets. So there is detail at all scales. True, some of the detail in the photo will be genuine “signal” and some will be noise, but I can’t tell from the photo how much we have of each.
The blue sky is a better guide to noise because it has no detail, so any small-scale variation must come from noise. Yes, I can see noise there, especialy when I zoom in x10, ie each pixel occupies 10x10 pixels on my screen. The noise in the sky is very much less than the variation in the sea. I would call it too minor to bother about, but I realise some folk would prefer it smoothed away. In that case, I suggest tweaking your denoiser until the sky is as you like it, provided it doesn’t noticeably change the sea etc.
As others say, light is a noisy thing, so photos even from ideal cameras will show noise in their photos. In general, larger sensels (ie the pixels on the sensor) should be less noisy than small ones. Cameras also introduce noise, but I don’t know how bad your camera is for this.
EDIT to add: I distinguish between noise and genuine detail, and I discourage you from accidentally removing detail. But that is a personal, aesthetic matter. If you prefer the image without high-frequency detail in the water, that is your choice.
Thanks for the recommendation; the gap between the different cameras I compared (e.g. the 40D which is similar to mine and the 90D) was not as large as I was expecting.
You’re right, both are underexposed:
sky_noise:
water_noise:
In fact, here’s the histogram and visible noise before any adjustments:
When compared to another photo taken from that same day where the histogram is much more evenly distributed to begin with (below), the difference is pretty obvious; I do NOT see the same kind of noise in the sky in this shot:
So likely the noise will be worse (or more visible) if underexposed?
@kofa good idea but defringe didn’t seem to help in this case.
That makes sense, but what doesn’t seem intuitive to me is that the variation would include obvious patches of red and green; I would expect variations of blue but not these other colors:
However the color picker says this is actually really blue, so maybe it’s just an illusion I see when looking at the image on the screen:
Well you have three colour channels, each with independent noise (in theory, demosaicing messes this up a bit).
And the sky isn’t pure blue, so there is signal in the red and green channel. That means some pixels will have more red/green and less blue, which means a colour shift towards red or green (magenta/cyan).
Also, as the signal in the red and green channels is much lower than the blue for the sky, the signal to noise ratio will be lower (more apparent noise).
And yes, underexposure will give more visible noise: you have to amplify the signal, which means you also amplify the noise (by the same factor). And your signal to noise ratio is lower when there is less signal (both through a larger relative contribution of read noise, and because a light signal has intrinsically a lower S/N at lower signal). That is the whole reason for “exposing to the right” (where you have to be very careful not to clip any colour channel…)
Yep, looks like that’s your culprit. @rvietor explains it well. Noise is proportional to the amount of light (signal) collected. The more light, the higher the signal to noise ratio, the less visible the noise. ISO simply amplifies whatever signal you have collected. The reason high ISO is associated with high noise is because it is typically used in low light situations, where there is less light, worse signal to noise ratio, thus more visible noise, which gets amplified. But you can get noisy images at low ISO if you don’t capture the light (underexpose), and you can get clean images at high ISO if you do.
However the color picker says this is actually really blue, …
Yes. In addition, the sea surface isn’t a pure mirror, so we don’t just have reflections of the sky. We also have refraction through the water, and some pixels will show light reflected from whatever is below the surface: seaweed etc.
Thanks for the additional explanations on how underexposure can contribute to more noise; I’ll plan on trying to overexpose (while watching the highlights) in the future to improve the signal to noise ratio
