Mmmm, looks interesting.
Any idea when there will be a Mac build. as I’m still on 5.11 public release.
Here’s hoping it’ll be soon.
I’ve experimenting the Capture Sharpening denoise tool with 3 or 4 somewhat noisy images and when comparing with/without pre-sharpening denoising, I found myself just trading sharpness for fewer noise. I would expect some kind of “remove noise while retaining sharpness” magic. Is this a overly-high or unexpected expectation?
I do find high-frequency areas receive more sharpening with pre-sharpening denoising enabled (by looking at and comparing contrast masks), but the difference in real images are less significant than what contrast masks show.
Denoise is probably the most difficult problem to solve in image processing. I don’t think there are any miracle solutions.
As I’ve already said, this “Pull Request” is an opportunity to test new experimental processes, both for noise and sharpening.
If we examine the first module, “Capture Sharpening,” it generally works well; the algorithm is very efficient. However, on some noisy or very noisy images, it doesn’t work; the mask remains uniform. This is due to the noise present.
The “Presharpening Denoise” function aims to enable the use of Capture Sharpening; for this, I slightly denoise the image beforehand. Keen to improve, I added the ability to perform this denoise not just with a small median, but with Wavelet acting on all three RGB channels. The system is more progressive. In my opinion, with few exceptions, you should limit the slider to a value slightly higher than the one that allows the mask to work. Beyond that, though it’s a choice, we apply denoising to the entire image; this can be useful in rare cases.
The ‘postsharpening denoise’ function is not intended to replace the other Denoise functions present in Rawtherapee, but rather complements them, particularly since at this stage (Raw), the GUI is not sensitive to pipeline malfunctions (preview/output difference). This will allow us to differentiate the action between structures and the background (flat areas), and, for example, denoise the background without (too much) affecting the main subject.
I attach the 2 new tooltips - written in my bad English , for your comments and changes in form and substance
Presharpening denoise:
In order for Capture Sharpening to work on noisy or very noisy images, applies either : a) Gradually by successive iterations a light ‘3x3 soft’ to a fairly strong ‘5x5 strong’ median denoising filter before Capture Sharpening. b) Wavelet denoising which has a different, more progressive action, mainly on the first 4 levels of decomposition(high and medium frequency). This helps clarify the contrast mask and allows Contrast Threshold to work, and thus Capture Sharpening to work properly. This also helps to reduce noise in the output image. Be careful not to use values that are too high so as not to degrade the image. The value that allows the mask to appear (or a little more) is usually sufficient.
Postsharpening denoise:
Allows you in raw mode, to do an initial denoising after Capture Sharpening taking into account the mask information. Its main use is to separate the action between structures where the noise is barely visible and the often plain background (flat areas) where the noise is very visible. Be careful not to use values that are too high, you can fine-tune the noise reduction later.
New executables are available
Text captur-noise
Jacques
2 Likes
Thanks for the explanation Jacque, will try your new build in my spare time. Looks like I need to try this on some really noisy images that traditional capture sharpening doesn’t work well.
I call that a manual in a tooltip! 
3 Likes
I think I found one image that works with it well 
Capture Sharpening enabled, presharpening denoise set to 21 (things start to change at 22, so 21 is the biggest value before any effect appears and essentially the same as 0):
Compare it to presharpening denoise set to 22, the noises in low-frequency areas are removed, while the edges of bike receive proper sharpening (see below).
Take a look at the sharpening mask of the post-denoise image:
And the “before” version for your amusement:
(Yes, it’s just a big white image.)
This image has relatively evenly distributed luma noises and low level of chroma noises, and the tool works well.
I tested this tool with a fairly well-exposed night shot (with a flash) a few days ago, I didn’t see any real changes since the front of the scene is noise-free and the back is completely dark, so no room for denoising tool to show its effect.
I also tested on an underexposed image (by 5-stop), it doesn’t work since there are too many luma & chroma noises (to be expected, it doesn’t have details to begin with).
So this tool is indeed awesome on certain kind of images!
Edit: It’s bad not to do proofreading when you’re not a native English speaker and when it’s 12 am…
1 Like
I just did various manipulations. The first one merges with 5.12, which brings the same functionality.
Even though it is not of direct interest to the user, I made some code improvements suggested by Copilot (Github), notably by replacing float** pointer variables with arrays or implementing std::unique_ptr for CurveEditorGroup in denoise. This makes the code more robust. Some parts of the code were (very) old.
I also made changes to the default values of vanishing moments used by the Daubechies wavelet decomposition (wavelet levels, Selective Editing, Denoise…).
They were defaulted to D4, I changed it to D6 which requires a bit more resources but gives a cleaner decomposition. In some places - for example PostSharpening Denoise, I even used D20 which gives a bit cleaner result on flat areas for denoise…Just a little more resources (minimal) and processing time (minimal).
I have also completed, due to the lack of documentation, more descriptive tooltips which are in fact a manual in a tooltip.
Executables
captur-noise
jacques
3 Likes
Quick question, when does b) wavelet denoising at first 4 levels kick in? Do I need to use a high “pre-sharpening denoise” value or it’s always in effect? And is it possible to let user to manually adjust denoising like how you do it in Wavelet’s “denoise and refinement” module (I was thinking of if doing it manually can help, since I think “denoise and refinement” is a powerful denoising/sharpening tool, but I just assumed it is a silly idea, so I didn’t bring it up weeks ago)?
As always, thanks Jacque for your development efforts and insights!
The presharpening denoise module is only intended to enable Capture Sharpening to work. I’m not saying it shouldn’t be used in other cases, but it acts on all three RGB components, i.e., on both luminance and chrominance noise.
It intervenes very early in the pipeline. It acts on the first 7 levels of decomposition from 2x2 to 64x64 with, as the action on the slider progresses, an effect first on levels 0 and 1 (2x2 pixels and 4x4 pixels) and little effect on the others, then on the higher levels.
Postsharpening denoise also acts very early on - just after Capture Sharpening. Its main purpose, as I said earlier, is to be able to use a mask that is identical between the preview and output, and therefore allow the plain background to be denoised from the rest of the image
It also acts on the first 7 levels of decomposition without differentiating between luminance and chrominance noise. Only the last level (64x64 pixels) is weakened in its action.
It is difficult to compare the action of these “Capture sharpening” modules with others, which separate luminance and chrominance noise.
Quietly repeat myself, these two modules do not remove or replace the other noise reduction modules but provide additional functions that were very difficult, if not impossible, to resolve without them.
But after examination, it’s not possible to do as in wavelet levels due to the position in the pipeline (we are before the assignment of a working profile), so we cannot - unless I am proven otherwise - adjust luminance noise and shapening in a differentiated manner. It can only be global.
In all cases, whether for presharpening or psotsharpening, you must use the minimum settings. In most cases, if Capture Sharpening works without it, you should not use presharpening denoise.
Thank you for testing
Jacques
1 Like
Understood, I learnt something new regarding how the tool works. I should find spare time to really read through your previous posts.
Not surprised my proposal doesn’t work due to technical limit, but that’s worth a try.
Thanks for your reply!
1 Like