Example image (Nikon D5200 NEF, 1/100s, f/2.8, ISO 1600) with same in/output and lens profile, chrominance noise reduction 25%. 2x magnification
RawTherapee 5.6 (LMMSE interpolation, 2 steps):
Lightroom 5.5:
As you see, the Lightroom denoising is much better. Many single dark pixels in RT while pixels are treated like a continuum in LR.
Using more LMMSE enhancement steps does not improve the result. LMMSE is still the best interpolation algo for high ISO images I find in RT.
(Here RT adds a hot pixel (only to the preview) where there is none.)
What can be done to come closer to LR? Line noise, hot/dead pixel filter and wavelet settings don’t help.
This is the original NEF file for download: _DSC0986.NEF (26.4 MB)
While Lightroom does a few more things under the hood than just demosaicing, it would be a good idea to post your raw file so people can play with it and maybe find a good solution for you
1 FLOSS apps generally expose more of the processing. This is advantageous because with patience we can do better and more informed processing. The purpose of using LMMSE is just that. It is careful not to bulldoze noise to the point where it mixes with signal, which would make denoising more difficult in the long run.
2 I find that the hot, dead and everything in between pixel correction in dt and RT are not aggressive enough in some cases. In that case, I just use G’MIC or GIMP to fill in areas affected by such pixels. I guess the most correct method is to find their coordinates and filter or mask them prior to the demosaicing step.
I’m not used to try finding a look similar to another software, but I’ve done my best this time.
I’m adding the LR version again just to look at both results one next to another, although the ideal scenario would be to open both images each on its own browser tab, and quickly switch between them to see the differences.
Lightroom:
RT 5.7:
The RT image is a bit darker, but that goes with one’s taste (it’s an image taken at night, after all).
I’ve used VNG4 as demosaicing algorithm (unusual, but works better in this image)
I’ve denoised the image just a bit. Only to get a similar result to LR
The hot and dead pixels have been applied to the max (the lowest threshold)
I’ve applied a bit of blurring with wavelets
Added some vibrance, too (again just to get a similar result to LR)
And tweaked a bit the exposure (not much, as is a night shot)
As @afre has told you, with RT (and other FLOSS apps) you have more control over the image, instead of getting a cooked image that won’t offer you what your sensor is capable of.
I would put it this way. LR and friends use recipes that may or may not be good for you but you have no say on what it does. In the end, comparison is moot. What you do with your tools is more important. It is like complaining about your height and comparing it with other people. You could do a number of things to make yourself appear taller or shorter (in short of extreme measures) but when you go to the doctor he or she would want to assess your true height.
Details: lost. I don’t want to get smoothness by paying with information. The luminance noise filter is usually not a good choice. My comparison is not pointless but important because denoising is a very sensible setting. Many smartphone images are destroyed by aggressive builtin algorithmic image “enhancing”.
@XavAL Thank you for your example! It was of course not my intention to create a “Lightroom look”, but I want to reduce noise that is there due to sensor’s limitations. I still prefer my profile to yours because the details in yours get “smoothed out” a bit too much. I can see it when loading your profile: A wave has gone through my image and taken some information with it I like the “true” luminance noise of an image to the degree where I cannot reduce it any more without loosing details. Probably I will soon understand more about wavelets and will find a better compromise between denoising and conserving information.
That is precisely why the way this was executed is pointless. There is no detail in the tiny LR crop to begin with. The location of this sample in the raw file is unknown. The latest version of RawTherapee was not used, the version of Lightroom is unknown. If you’re going to stir the pot, do it properly. A proper comparison would compare latest versions of both programs, provide full settings used (PP3, XMP, etc), isolate the processing to only denoising and nothing else, use several samples which contain areas of fine detail as well as low detail, luminance and chrominance noise as well as luminance and chrominance detail, compare the effects of denoising on an image which has both light and dark areas to see whether denoising is uniform or whether it improves one at the cost of the other. Then only can we begin to draw conclusions.
Well, I didn’t really like my processing, either.
Now that you have told us what you really are looking for, I have processed the image my way. Well, in fact is not fully processed, but instead just a starting point, so you can go wherever you want with it.
You may want to take a look at this noise reduction challenge thread, so you can see how much you can reduce noise, but not without generating artifacts or loosing something along the way.
This is the new profile: _DSC0986-2.png.out.pp3 (11.4 KB) (I’ve used a custom input profile made for my D5100 that I can post if you need it)
I’m not saying that you will like my current processing, but just that you will have another starting point with it.
A couple of comments:
the Noise reduction tool is activated to just auto-remove chroma noise
the noise reduction in the Wavelets tools has been applied to reduce the contrast of the very finest details (not really a noise reduction), to hide a bit the darkest pixels
I didn’t look anywhere in the image but the crop you posted, so I missed it. It was due to fringing in the image. Now it is corrected in the current profile, and you can see the result:
P.S.: if you want to start using the denoise module in the Wavelet tool, you can take a look at this Rawpedia explanation about it. It’s written in Spanish, but you can easily translate it with Deepl, Google translator, or the online translator of your choice.
I tried to compare RawTherapee 5.7 vs Lightroom 9 ( which is their latest build ) , and I prefer the result I’m getting from RT. I had to be modest with Lightroom NR in this picture (Luminance 30).
Here is the result from Lightroom and DxO
I like the “true” luminance noise of an image to the degree where I cannot reduce it any more without loosing details. Probably I will soon understand more about wavelets and will find a better compromise between denoising and conserving information.
