Available as a plug-in for Adobe Photoshop and as a process for PixInsight, NoiseXTerminator can eliminate unwanted noise while leaving stars untouched.
Noise is the bane of all astrophotographers. We struggle to record the faint signals from deep-sky targets that can become lost in a blizzard of noise without best practices at the telescope. We use low-noise or cooled cameras to take multiple images, and stack those in processing to average out noise.
But we also apply noise reduction at the computer, using software filters and programs, though not always with the best of success. The software often has a hard time telling noise from stars, so smoothing noise also wipes out stars and subtle details.
To address the issue, astrophotographer Russell Croman, under his RC-Astro brand, has introduced NoiseXTerminator. It is one of a new generation of artificial intelligence or “AI-powered” noise reduction programs that use neural network training on thousands of images to better tell unwanted noise from wanted content. But unlike mainstream programs such as Topaz DeNoise AI and ON1 NoNoise AI, NoiseXTerminator has been trained specifically on astrophotos, where stars can look like noise.
The bottom line: It works remarkably well, smoothing out even extreme noise, all while leaving the faintest and smallest stars alone, or sharpening them, if desired. I think NoiseXTerminator just might obsolete all other noise reduction routines and programs for deep-sky imaging.
I primarily tested NoiseXTerminator as a plug-in filter for Adobe Photoshop (it requires the latest Creative Cloud version). It does not work from within Adobe Lightroom.
It is also available as a process for PixInsight (both Mac and Windows versions), where it requires at least v1.8.8-9 of PixInsight.
NoiseXTerminator will also work in Serif Affinity Photo, for those who prefer that Photoshop-like layer-based program. (I checked it did indeed work, as many Photoshop plug-ins crash Affinity, despite appearing in Affinity’s Plug-Ins menu.)
At the time of my testing in May 2022 the Photoshop version was very new, at v0.9.1, while the PixInsight version was at v1.0.0, though with both using v2 of the AI module. As with Croman’s other AI-based program, StarXTerminator (which I reviewed earlier for AGT), the AI module gets updated regularly to improve performance, an upgrade which is separate from the program itself. The program’s dialog box provides a button to download the latest AI module. The PixInsight version allows choosing which AI module to use; the Photoshop version does not.
TESTING ON DEEP-SKY IMAGES
The other adjustments are simple but effective: one slider varies the Denoise level from 0% to 100%, while the other sets the Detail from 0 to 100 (both given as 0.0 to 1.0 in PixInsight). The Detail slider performs an AI-based sharpening that can bring out faint stars without increasing the visibility of fine-scale noise. That worked very well.
However, as the instructions advise, turning up Detail too high (much above about 30, I found) begins to introduce unsightly dark halos. By contrast, the Denoise setting seemed best when set high, to 80 or 90. However, your mileage will vary depending on the image’s size and content.
The Photoshop version works as a smart filter when applied to a smart object, allowing the filter to be re-opened at any time and its settings changed. However, the version I tested remembered the settings that were last used, which might not have been the settings used for that image. That’s contrary to how smart filters should work. They should always reopen with the settings applied to that specific image.
The PixInsight version offers the option for applying NoiseXTerminator to unstretched linear files, the recommended practice, to eliminate noise before applying curves and other forms of “histogram transformations” that will permanently stretch the image and enhance contrast. Doing so will make any noise more apparent, so it’s best to smooth it out before boosting contrast.
Applying NoiseXTerminator to large 45-megapixel images was fast, even on my 5-year-old Mac, taking no more than 20 to 30 seconds in either Photoshop or PixInsight.
I found the effectiveness of NoiseXTerminator varied considerably depending on how much noise was in the image. When testing it on single images riddled with noise, it provided an amazing level of improvement, as shown above with the PixInsight example of a galaxy field, where faint outlying structures became visible.
But on stacked images that were already low in noise, the program produced only a subtle level of improvement, as one might expect. In the above example, the old Reduce Noise filter long a part of Photoshop did almost as good a job as NoiseXTerminator for smoothing the last bit of random background mottling.
Also, as remarkable as NoiseXTerminator is, it does not eliminate the hot pixels from thermal noise that can pepper images. Indeed, by smoothing the overall background noise, NoiseXTerminator can make those random hot pixels stand out even more. An option for eliminating outlier hot or dark pixels as a form of intelligent cosmetic correction would be welcome.
Such thermal noise is still best eliminated by the application of good dark frames (or with DSLRs and mirrorless cameras, by turning on Long Exposure Noise Reduction). Using “dithering” at the telescope to shift each frame at random by a few pixels (done by the auto-guiding software) can also reduce hot pixels when stacking and aligning sub-frames.
However, with NoiseXTerminator working as well as it does (and it should improve with future AI updates), I can see astrophotographers deciding they don’t need to shoot as many sub-frames, or realizing they can now rescue old images they had abandoned as being too noisy to use.
TESTING ON NIGHTSCAPES
I was curious to see how well NoiseXTerminator would work on nightscape images, where we often contend with high-ISO noise in the shadows of foregrounds lit by moonlight or starlight, and yet where the landscape can have lots of fine detail.
As I show above, the settings that work well for deep-sky images, such as 90% Denoise and 20 Detail, ruined the foreground, producing an overly smooth, plastic look lacking in sharpness. Boosting the Detail only yielded harsh, artificially sharp edges. While backing off both settings did work, I found the result was still not as good as using mainstream AI noise reduction programs.
I reviewed Topaz DeNoise AI previously for AGT. I found that while it does work very well on “normal” scenes, it can produce unusual patchy artifacts on star fields and deep-sky images. In this case, its AI training knows about earthly scenes, but not about stars, just the opposite of NoiseXTerminator.
ON1 NoNoise AI has been my favorite noise reduction program of late, as it can produce very good results on both earth and sky components, but only if its default settings are backed off a lot, especially its Sharpening and Enhance Detail sliders, to avoid unsightly halos.
In fairness, NoiseXTerminator is not advertised as being suitable for other styles of astrophotos. So don’t expect it to work well on nightscape photos, except perhaps if applied just to the sky layer.
But for the specialized demands of deep-sky images, I can highly recommend NoiseXTerminator. The same $60 purchase price and license allows you to install it in both Photoshop, for use later in a processing workflow, and in PixInsight to apply early in a workflow. Download the trial copy first to see if it works on your computer, and fits into your image processing routine.
Plus: Superb noise reduction; easy to use within Photoshop or PixInsight
Minus: Works best only on deep-sky images; not suitable for nightscapes
Price: $59.95 ($39.95 for owners of other RC-Astro software); Free 30-day trial copy available
Systems: Windows 10 and later; MacOS 10.15 Catalina and later (including native support for Apple Silicon M1); Linux Ubuntu 18.04 and later