Yeah, it was kind of an accidental discovery that somehow works very well. You can also use it with very low ISO images: by setting the detail slider to around 90, it can fully eliminate the subtle luminance noise in the smooth parts while keeping nearly all the details intact. I find this works much better than setting luminance to something like 5~10 and keeping details to 0, which is not strong enough to remove noises in the background and is already starting to destroy details.
I agree with all you say. Also, I’d like to add that I much prefer the rgb method to the default lab one. Not only it does a much better job (IMHO) at preserving colours, but it also allows to be less aggressive with nr in general, as the grain pattern it produces is much more pleasing (or better, much less disturbing 
than what lab gives. Has anyone had similar experiences? Perhaps it depends on the camera as well?
I already noticed that !
Try on a noisy photo of Imaging Resource’s still life … RGB denoise (EDIT: should) destroys the red textile while lab keeps some detail …
EDIT … I had not tried with recent RT … just remembered from the old RT forum … the reason for addind the Lab mode was that RGB mode destroyed the color … then Jacques added Lab mode which was better with colors
But looks like it’s the inverse now !!!
@agriggio Interesting. I made a comparison with this snapshot of my doggo I took tonight. This was shot on the Sony A6000 at ISO 6400, demosaiced with AMaZE and 3 steps of false color suppression, impulse NR set to 65 and used high quality automatic global for chrominance denoise.
Lab*, 100/30
RGB, 100/15 (lowered the detail slider to match the overall noise level)
The Lab* one seems to produces an image of higher contrast and if you zoom into 100%, you should see that Lab method also produces much finer noise patterns. However, the noise from RGB actually doesn’t look that bad when zoomed out, thanks to less contrasty shadows that help to hide these noises. Lab’s higher contrast also makes the fur look sharper.
Now onto colors. The first thing I noticed is that Lab* method tones the darker areas with an unnatural red, which I find very distracting. The RGB method on the other hand makes highlights a bit more saturated, which I think works well for this image. The main issue with RGB here is that it produces messy color blobs around the color transitions in the background, which is only barely visible in Lab method. Perhaps increasing the chrominance NR slider could help migrating the problem.
Hi @PkmX, thanks for the analysis! One more thing I discovered is that I prefer the results I get when using the “Standard” quality of NR instead of “High”. Although the latter does certainly a better job at reducing color noise, the drawback IMHO is that it “dampens” the colors too much, resulting in a “dull” image. See the two examples below (I downscaled the pics because I’m on a slow connection, but the difference should still be visible).
standard quality
high quality
＠agriggio Yes, I’m aware of that difference. That’s why in my reply above I suggested lowering the chrominance curve by a lot, which can recovery some of that colors lost without reintroducing noise. This works very nicely for low ISO images.
I think I should also take the opportunity to showcase how well RawTherapee can deal with extreme noise. Here is another image of the black dog that was shared in my PlayRAW thread, and this time it is shot at ISO 25600 on the Sony A6000 (APS-C), the maximum ISO allowed on the camera. The only light source was a street lamp about 7 or 8 meters away, and the subject being a black dog means that I had to crank the ISO all the way up to the maximum.
RAW Image
Eww, awful noises everywhere.
SOOC JPEG
Lots of details are smeared on the fur and there are still loads of noises visible in the background.
RawTherapee
All colorful noise are simply gone. The noise pattern in the background is fine-grained and unobtrusive, and nearly all details in the fur are still intact. This is seriously impressive in my opinion.
OK guys, how well could you process this image:
_IMG5577.DNG (12.8 MB)
I couldn’t do better than this:
@Trickortreat I think the main issue is that there is either motion blur or the town is slightly out of focus. There actually isn’t much noise in the image.
Here is my quick attempt:
By the way, the auto levels function in RawTherapee is totally messed up with this image, it sets lightness to some value >90. I guess this is due to the excessive amount of black in the image?
But your dog is the wrong colour and the grass has gone from green to orange-ish! I think the jpeg is a better record of the dog.
The color difference is caused by me using a different color profile, as well as changing the white balance plus some H-H adjustment in Lab* tool and finally applying a film emulation CLUT. With all these adjustments turned off, it looks fairly close to the RAW image:
Could you post your pp3, i like how you extracted more detail than me.
Center point is in focus, other is some lens issue because at f2.8 corners are somewhat soft and darker + there is noticeable field curvature
Yep - i got that lightness issue too, just dialed it down manually to 9
@Trickortreat Here it is: _IMG5577.DNG.pp3 (9.5 KB)
You need Kodak Portra 400 NC 1 -.png for film emulation and Rec2020-elle-V4-g10.icc from elle’s ICC profiles to achieve the tonality and colors. I then used the wavelet tool to bring back details.
one more thing. what effect do this tools have on noise:
- local contrast in shadows/highlights
- and sharp mask under same tab
tnx
yes, a lot better. I’d like to understand more about the noise reduction you did here. Would you mind posting up your .pp3 for this one please?
Here’s my attempt at it.A bit on the darkish side. Problem with this image is that you need some pulling up of mid and lower mid tones which induces more noise that can be troublesome to get rid of…
Deepest blacks are best left completly crushed to the left side of histogram