As we know, some noise reduction in the linear stage is critical for a successful image. I have been using TGV DeNoise with a low contrast mask. I am interested in learning what methods others use and their experience with them.
PixInsight Linear Noise Reduction
Well Written Engaging
I have three methods which I use to different extend but always in this sequence:
- TGV DeNoise with a support based on the "luminance" channel* (L from RGB if image is RGB otherwise the grayscale image itself). Don't know if you meant mask or support in your above post.
- MultiscaleMedianTransform protecting brighter areas through an inverted "luminance" channel* (def. see above).
- Denoise in AffinityPhoto on the linear and also non-linear image. For a HT, I am using AffinityPhoto and so I can dynamically adjust the Denoise parameters. Sometimes I mask this Denoise filter, sometimes I don't.
* "luminance" channel is usually a HT of the linear image with some possible curves adjustment. I'm usually blurring the mask through a convolution filter to smoothen the mask.
It really depends on the raw data which process I use and to which extend.
Björn
- TGV DeNoise with a support based on the "luminance" channel* (L from RGB if image is RGB otherwise the grayscale image itself). Don't know if you meant mask or support in your above post.
- MultiscaleMedianTransform protecting brighter areas through an inverted "luminance" channel* (def. see above).
- Denoise in AffinityPhoto on the linear and also non-linear image. For a HT, I am using AffinityPhoto and so I can dynamically adjust the Denoise parameters. Sometimes I mask this Denoise filter, sometimes I don't.
* "luminance" channel is usually a HT of the linear image with some possible curves adjustment. I'm usually blurring the mask through a convolution filter to smoothen the mask.
It really depends on the raw data which process I use and to which extend.
Björn
Previously, I would go through the more labor intensive masked MLT process in the linear phase, but for several months have had excellent results using Russ Crowman's excellent NoiseXTerminator process in both the linear phase and then right after stretching again. The default of 0.9 is way too strong, so I trial different levels in preview windows, and typically end up @ 0.4-0.6 for linear images (less for drizzled images, more for undrizzled images), and then again at even lower settings right after the initial stretch.
Helpful
Thanks, Bjorn and Rob, for sharing.
When I first started out, I'd use MLT with a linear mask, but then moved to the TGV DeNoise approach with stretched luminance support and low contrast mask (the Rista method). This served me well for a number of images. But recently, I've been struggling trying to balance reducing noise uniformly in the shadows versus completely eliminating it in certain areas resulting in an ununiformly blotchy result. I generally like to leave some graininess in my images.
I have tried NoiseXTerminator as well but it seemed to give me less control than TGV DeNoise. It certainly is a nice piece of software though, and its ease of use cannot be matched.
When I first started out, I'd use MLT with a linear mask, but then moved to the TGV DeNoise approach with stretched luminance support and low contrast mask (the Rista method). This served me well for a number of images. But recently, I've been struggling trying to balance reducing noise uniformly in the shadows versus completely eliminating it in certain areas resulting in an ununiformly blotchy result. I generally like to leave some graininess in my images.
I have tried NoiseXTerminator as well but it seemed to give me less control than TGV DeNoise. It certainly is a nice piece of software though, and its ease of use cannot be matched.
Well Written Insightful Respectful Engaging Supportive
I apply linear TGVDenoise with a slightly darkened luminance mask (applied inverted for the background), with local support via a traditional star mask (created from a non-darkened luminance mask). I often come back to TGVDenoise in non-linear processing (same inverted, darkened luminance mask, but without local support) if I feel the need. This said, there's no substitute for exposure time when it comes to noise reduction! I've found that increasing the number of subs often negates the need for non-linear NR; linear NR often is sufficient.
I'll note that right before the non-linear stretch, after TGVDenoise and any applied decon, an application of MLT in 2 phases (lum, chrom) is very helpful to smooth out the background.
I'll note that right before the non-linear stretch, after TGVDenoise and any applied decon, an application of MLT in 2 phases (lum, chrom) is very helpful to smooth out the background.
Helpful Insightful
Doug Summers:
I'll note that right before the non-linear stretch, after TGVDenoise and any applied decon, an application of MLT in 2 phases (lum, chrom) is very helpful to smooth out the background.
Good point. I forgot to mention in my post:
I often use the Denoise tools in PI in the Lab mode instead of RGB to treat luminance and chrominance differently.
Well Written Insightful Respectful Concise
Get the NoiseXTerminator and be done with it.
Lately I have been using a touch of TGV in linear state to reduce small scale noise. Then a round of NoiseXTerminator later in processing. The exact settings and order changes from image to image.
I use Jon Rista’s method https://jonrista.com/the-astrophotographers-guide/pixinsights/effective-noise-reduction-part-1/
Fiddly but works beautifully
Geoff
Fiddly but works beautifully
Geoff
Geoff - yes, agreed. I've been using Jon Rista's method too. It works great. I've tried NoiseXterminator, which is easy, but doesn't give me nearly the control that properly masked and supported TGVDenoise does in linear state. I guess I was looking for a magic solution that would avoid the fiddle, but there isn't one. You've got to put in the work, I suppose.
Well Written Insightful Respectful Concise Engaging Supportive
You can mask NoiseX to your heart's content in exactly the same way your mask MMT and TGV.
You can mask NoiseX to your heart's content in exactly the same way your mask MMT and TGV.
No doubt, but if I am going to do that, what distinguishes it from the available packages in PixInsight, especially if I need to pay for it? Things like TGV DeNoise and ACDNR work very, very well once you learn how to use them, so the use case to me for something different is speed and simplicity. I don't mean to rag on NoiseXTerminator, which I think is a fine piece of software that will be improved even more over time, only making the point that if I need to do the things with it that I need to do with TGV DeNoise etc., to get it to work for me, then I might as well use the tools I am familiar with. I realize other people may legitimately come to different conclusions and I accept that as a valid decision.
Helpful Insightful Respectful
Even with a mask NoiseX is substantially less work. But yes I agree that if your skilled enough the built in tools can work well.
At least in my case, since NoiseXterminator I have not touched any of the other PI noise reduction tools.
As close to “magic” as you can find.
As close to “magic” as you can find.