NoiseXTerminator vs MureDenoise

[STEVE333]

This is a comparison of NoiseXTerminator (NXT) vs MureDenoise in PixInsight. Because I'm doing NB imaging all of my images are monochrome. I haven't tried it with color images.

I have been using MureDenoise for some time now to perform noise reduction for my linear images because, up till now, it has donr a better job of noise reduction than I can do any other way. Now I want to test NXT.

The pictures below show a comparison between the two approaches for a linear image for both unstructured background and for target details. I used NXT with Denoise = 0.90 and Detail = 0.15, the default settings.

The pictures on the left show a zoomed in view of the original stacked image.
The middle pictures show the MureDenoise processed images.
The rightmost pictures show the NXT processed images.

UNSTRUCTURED BACKGROUND

DETAILS

I have found NoiseXTerminator (NXT) to be far superior to MureDenoise:

• Visually the NXT processed images are much less noisy than the MureDenoise processed images.
• Many of my stacked images have significant noise but NXT has been able to smooth it out every time.
• NXT runs faster than MureDenoise.
• NXT has not left any artifacts (cross-hatched regions) as MureDenoise sometimes does.
• NXT has also done a fine job or retaining the details.
• NXT seems to sharpen the stars a little.
• The amount of noise reduction can be adjusted with NXT, not true with MureDenoise. Very helpful!! The picture below shows examples of two NXT noise reduction levels. The image on the left is the original image. The middle image has Denoise = 70 and the rightmost image has the default setting of Denoise = 90.

I used NXT on my recent mosaic and was completely satisfied in all aspects. It has made my processing task much easier.

Hope this is useful. I'll be glad to answer any questions if I can.

Steve

 

[rwatters]

Thanks for posting this analysis Steve. I am also trialling NoiseXterminator currently and it looks to be a useful tool. Based on my usage and your analysis above, I am still keeping an open mind though. Looking at your test images for example I may challenge the conclusion that NoiseXTerminator has retained the same detail as MureDenoise. However, we must be very careful in that there are settings in both tools that we could adjust to produce different results so I don’t think that we can lock in the conclusions just yet. For example, the amount of noise reduction can be adjusted in Mure Denoise with the Variance Scale parameter. NoiseXTerminator is certainly an easy tool to use as it doesn’t require the prerequisite of working out the detector settings first (though this only has to be done once for each camera).

 

[STEVE333]

Thanks for posting this analysis Steve. I am also trialling NoiseXterminator currently and it looks to be a useful tool. Based on my usage and your analysis above, I am still keeping an open mind though. Looking at your test images for example I may challenge the conclusion that NoiseXTerminator has retained the same detail as MureDenoise. However, we must be very careful in that there are settings in both tools that we could adjust to produce different results so I don’t think that we can lock in the conclusions just yet. For example, the amount of noise reduction can be adjusted in Mure Denoise with the Variance Scale parameter. NoiseXTerminator is certainly an easy tool to use as it doesn’t require the prerequisite of working out the detector settings first (though this only has to be done once for each camera).

Good points. Thanks for the constructive feedback.

Steve

 

[StuartT]

This is a comparison of NoiseXTerminator (NXT) vs MureDenoise in PixInsight. Because I'm doing NB imaging all of my images are monochrome. I haven't tried it with color images.

I have found NoiseXTerminator (NXT) to be far superior to MureDenoise:

I have not used either. I always use EZ Denoise. Can you comment on how NXT compares to that?
Thanks

 

[Marcelofig]

I have not used either. I always use EZ Denoise. Can you comment on how NXT compares to that?
Thanks

I can comment. I mostly do NB and used to use EZ Denoise, but after installing NXT I have never touched another noise reduction tool. Amazing results literally in a couple of seconds.

It has a 30 day demo, give it a try.

 

[StuartT]

ok, I've done a test on a horribly noisy astro image and a terrestrial photo to which I'd added some Gaussian noise (not strictly fair on these tools, I realise, but interesting nevertheless).

I would say RC Astro's Noise XTerminator is superior to EZ Denoise and doesn't appear to be losing any detail.

 

[Buzz]

Not to bump the topic, but to add some further observations:
I concur with the findings - I was wondering about NXT and searching the forum, found this thread. I have run side-by-side between MureDenoise and NXT on wideband and NB stacks (>10hrs). NXT is cleaner, without any apparent downside and less destructive on faint structures. Running deconvolution and NXT one after the other, I found that I have better definition of faint stars, without ringing on some stars, if I deconvolve before running NXT (same settings in both case). I cannot think of a reason not to use it. I didn't just use default settings but backed off somewhat to find the right balance.
I also noticed, running DBE afterwards, that faint nebulosity in the background was less blotchy with NXT than MureDenoise.

 

[Buzz]

Russell - I sent you a PM about the suggested workflow of using SXT/NXT and Deconvolution in PI - might be useful to put it here, for other's benefit.

 

[JST200]

Russell - I sent you a PM about the suggested workflow of using SXT/NXT and Deconvolution in PI - might be useful to put it here, for other's benefit.

Yes, please!

I've just started investigating the value of NXT/SXT and been blown away at their effectiveness.

I've just been processing the Horsehead nebula, using 5 hours of subs using different exposures and light pollution filters. I could get a pretty good result using standard denoise (e.g . MLT, EZ Denoise) and star removal tools (e.g. StarNet, StarNet 2). But using NXT and SXT I have been able to pull detail and colours I had always thought were there, but could never get. What's more, what took a good four days using the standard tools took a couple of hours using NXT and SXT!

So I would truly appreciate any tip and hints available.

Thanks, Jim

 

[dld]

It is welcome to have a new denoising algorithm at our disposal but before drawing conclusions on which is best, don't we need some performance tests using some standard denoising metrics (PSNR, SSIM) from people who are proficient in both tools? Or at least a comparison like this good old tutorial?

PixInsight — A Comparison of Noise Reduction Algorithms

PixInsight - Advanced Image Processing. PixInsight is a modular, open-architecture, portable image processing platform available for FreeBSD, Linux, Mac OS X and Windows.

pixinsight.com

Sorry for being a skeptic but I am overwhelmed by the plethora of resources offering advice (?) on PixInsight (youtube and others) and I wish the PI forum to be a high-SNR resource and our last resort for the software we all love.

 

[Buzz]

I understand. I have been using PI for many years and publish books on Astrophotography, including large sections on using PI and a series of well regarded YouTube videos. I would say I'm proficient but not necessarily an expert as a few individuals are. I have been using MureDenoise since it came out to good effect. My aim is to make pretty pictures; I do use numerical SNR figures in some cases but it is the noise appearance in the final image that I'm concerned with. The two do not always tally (witness the endless discussions on LRGB vs RGB). I'm just saying that my first attempt using NXT was visually better than my careful use of MureDenoise. I can re-create the test, measure and post... but it was not a nuanced difference and numbers (IMHO) were irrelevant.

 

[Buzz]

Here you go - I just chose another stack at random. I notice that the differences between the methods are less with a lower noise stack. My first attempt used a wideband stack which had a lot more shot noise in it and the differences were more obvious.

The example below is with an Ha stack of about 40 frames. QHY183 CMOS. I processed the full frame and then cropped and applied the same stretch to each. I used the noise script in PI to measure the full frame noise. There are four: original, Mure (8 cycles), with 1.0 setting and 0.7 setting and NXT with 0.8 setting (0.2 sharpening).
SNRs for Orig: Mure1.0 : Mure0.7 : NXT(0.8/.2) are 0.4 : 0.5 : 0.25 : 0.09. respt.

 

[Buzz]

Out of interest I downloaded the zipped project files with the generated opposing gradients and added noise and created an NXT one. It did not do well. But then I remembered that NXT is adapting to the image elements, so this artificial target is perhaps less relevant to astrophotographic images. I was reminded of this discriminatory behavior when I ran StarXterminator on a stack of the Fireworks galaxy. It removed all the stars but cleverly left behind several small faint spiral/oval galaxies which I had not even noticed before in the dense starfield.

 

[Osteopatia Postura]

Russell - ti ho inviato un PM sul flusso di lavoro suggerito per l'utilizzo di SXT/NXT e Deconvolution in PI - potrebbe essere utile metterlo qui, a beneficio di altri.

Ragazzi sono appena incappato in NXT volevo provarlo,prima usavo MLT e poi in fase non lineare Tgvdenoise in fase non lineare. Usando anche la deconvoluzione dopo Spcc quando conviene usare il processo NXT? Grazie