Denoise challenge - raw, jpeg, tiff provided

Dear Fujians

I recently went for Neat Image and reported on a denoising experience. Here's my report: Denoising high ISO images // Test: darktable vs Neat Image (Linux user). The discussion went in the direction "Topaz DeNoise AI vs Neat Image." I did a direct comparison and have to say that I am impressed by the result from Topaz (demo version). Neat Image is better for my workflow since I'm on Linux.

Now I am still a novice with Neat Image. I played a bit with its advanced mode, I can produce good results. However, already with the standard procedure, Topaz does a really great job with Pablo's hair IMO. Also with the standard procedure, the background is less noisy with Topaz (a little nervous, though).

I am curious, what do you get with your way of denoising? Just post here please. You're of course free to use whatever software you want.

Please find the following image files in this folder.

• 20220714-Denoise_Test-4641.raf - the raw file
• 20220714-Denoise_Test-4641_darktable_export.jpg - the darktable export as a jpeg, not sharpened, not denoised
• 20220714-Denoise_Test-4641_darktable_export.tiff - the darktable export as a 16 bit tiff, not sharpened, not denoised
• 20220714-Denoise_Test-4641_OOC.jpg - the OOC image, NR set to -4, sharpening set to +1. Although NR was set to -4, there is clearly more noise reduction as compared to the darktable export where I deactivated noise reduction.
• The folder link is valid four weeks from today.

About the image

• It shows Pablo, one of our dwarf rabbits. He is about two years old. They're allowed to be in the garden during daytime.
• X-T4
• XF 50/2, aperture f/4
• ISO 12800
• Classic Chrome
• Focus on Pablo's eye

Neat Image v9.0.2 Samples

My first result with Neat Image. Beginner's mode. I do not like how Pablo's hair came out and the region around the eye, there are clear artifacts. You need to zoom in to recognize those artifacts.

Another try with Neat Image. Maybe almost 'overdone' but I like the hair and the region around the eye much better when zoomed in. Advanced mode, highest quality, noise level raised to 90 % (default = 0 %) , luminance level unchanged (60 %), no sharpening, detail recovery enabled with std settings. I learned when I want to improve Pablo's hair, I need to raise the noise level bar. When I want to have a smoother background, I need to enhance the luminance level but this is in turn detrimental for the rabbit's hair. To optimize both, background and rabbit, I'd need to work with layers which easily gets time consuming. Artifact removal does not change the picture much (it was activated, though). Same with edge smoothing (not activated). Detail recovery has a huge influence. Sharpening was detrimental (produced artifacts) so I did not apply it.

Topaz DeNoise AI v3.7.0 demo version sample

Topaz DeNoise AI demo version, standard procedure, no time invested to play with settings. I like the outcome. Background has very low noise but is a bit restless IMO.

Get - set - go!

Martin

Rightsaidfred wrote:

Another try with Neat Image. Maybe almost 'overdone' but I like the hair and the region around the eye much better when zoomed in. Advanced mode, highest quality, noise level raised to 90 % (default = 0 %) , luminance level unchanged (60 %), no sharpening, detail recovery enabled with std settings. I learned when I want to improve Pablo's hair, I need to raise the noise level bar. When I want to have a smoother background, I need to enhance the luminance level but this is in turn detrimental for the rabbit's hair. To optimize both, background and rabbit, I'd need to work with layers which easily gets time consuming. Artifact removal does not change the picture much (it was activated, though). Same with edge smoothing (not activated). Detail recovery has a huge influence. Sharpening was detrimental (produced artifacts) so I did not apply it.

Hi Martin, this is about the best Pedro I could manage with Lightroom/IXT + Topaz DeNoise. The background looks blotchier than I'd like, but the detail retention is good.

Lightroom/IXT + Topaz DeNoise A.I.

I always have a little LR sharpening and NR in the mix, but these were my Topaz settings (no masking)...

I almost always have the Remove Noise setting at zero in this mode (which isn't actually zero) , but this image was somewhat atypical.

If you're curious (I was), this version is just Lightroom, no IXT, and no Topaz or DxO etc. Softer, but not especially objectionable, IMO...

Old school with just plain old Lightroom

Fun and probably educational challenge Martin.

I liked Eric's idea to use native noise removal I decided to do the noise removal the old way.  I did two conversions.  First I pulled down the whites and highlights to recover some detail.  The I adjusted the white balance using the ACR sampling tool on the rabbit's white area.  Next I darkened the image till all the white detail was visible and completed the conversion by bringing the image into photoshop.  I created another layer making the first conversion the background layer.  I named the new layer noise reduction.  Then I opened the image in ACR again and adjusted noise removal to my liking which was a rather strong 80 and left everything else from the previous conversion and opened in Photoshop.  I copied the image with noise removal into the noise removal layer of the first image.  I selected the background layer and sharpened with Topaz AI Sharpen and then applied a bit more sharpening via USM.  I selected the top layer and erased the areas I want to reveal the sharpening and then flattened the image.  As I'd made the image darker to protect the whites I used shadow/highlights to taste and then checked in levels and brought up the white point.  As there were a few bright and distracting spots I cleaned them up via cloning and or the burn tool.

This worked so well it makes me wonder if I should contuse to pay for the plugins.  Will need to test more to see.

Morris

maltmoose wrote:

Capture one with tweaks, exported to topaz denoise low light mode.

Nice clean detail here.

I understand and appreciate the exercise of this thread, but I can't get past myself looking at sharpness for comparison. I don't mind noise so much, so I always feel that denoising itself seems to soften an image.

Am I alone in this?

Samuraidog wrote:

I understand and appreciate the exercise of this thread, but I can't get past myself looking at sharpness for comparison. I don't mind noise so much, so I always feel that denoising itself seems to soften an image.

Am I alone in this?

Yes and now.  Some denoise software removes noise where there is no detail and sharpens where there is detail.  I prefer to do this my self choosing where to do each which allows me to apply more noise reduction and sharpening as necessary.  Take a look above at the results I produced.

Morris

Really fun challange, here's my attempt using DXO.

Samuraidog wrote:

I understand and appreciate the exercise of this thread, but I can't get past myself looking at sharpness for comparison. I don't mind noise so much, so I always feel that denoising itself seems to soften an image.

Not so much with today's A.I. powered NR. My detail came out a little crunchier than I would have liked, but a lot of noise has been cleaned up with very little worthwhile detail being sacrificed in the process..

terraflop wrote:

DxO PL5 using DeepPRIME and a little WB / contrast tweak.

As these two aren't radically different in overall processing approach, I thought them good to compare. To my eye, Topaz puts in a pretty good showing (IMO), but by cleaning up the image before demosaicing, to my eye, DeepPrime definitely has a bit of an edge when it comes to separating the noise from the fine detail with very noisy/high ISO images like this one. I will have to take another look at PureRaw2,(although it's beyond annoying that it doesn't have the same fine-tuning capability as PL5).

Lightroom/IXT/Topaz DeNoise

Erik Baumgartner wrote:

Rightsaidfred wrote:

Hi Martin, this is about the best Pedro I could manage with Lightroom/IXT + Topaz DeNoise. The background looks blotchier than I'd like, but the detail retention is good.

I always have a little LR sharpening and NR in the mix, but these were my Topaz settings (no masking)...

Hi Erik

I'ce scrolled through all the results. IMO this is one of the best, if not the best, results.

Detail level is very good, and it is overall harmonic and feels authentic.

Thank you very much.

Regards,

Martin

Erik Baumgartner wrote:

If you're curious (I was), this version is just Lightroom, no IXT, and no Topaz or DxO etc. Softer, but not especially objectionable, IMO...

Fully agree. Still a lot of noise, detail level not good.

It's good to have benchmarks.

That's why I am so interested in this challenge. The image itself is certainly more on the challenging side, as I mentioned in the other thread, especially Pablo's hair.

Thank you.

Regards,

Martin

RetiringGuy wrote:

A quick try using Capture One alone. Auto white balance instead of 'as shot'.

RG

Thank you. I see artifacts, the reproduction of Pablo's hair is not very pleasing IMO.

Most importantly. nothing personal, this s about a technical challenge. I find the image challenging, and it is very interesting to see the different approaches.

Wrt the colors, yes I know, I am a bit special, I often like the more withdrawn and cooler colors, I love Classic Chrome.

Regards,

Martin