When I started processing astronomical images, I went with the common open source tools Siril and GIMP. Siril worked very well for calibration, registration, integration, and some basic post-processing. I would use GIMP to finish images as best as I could figure out. This combination couldn’t give me everything that I wanted to do with my images, so I eventually purchased PixInsight. The thing that drew me to PixInsight over other proprietary astronomical processing programs is that they actually had a port for FreeBSD. Well, that didn’t last long, as they withdrew support for FreeBSD mere weeks after I had purchased a license.
PixInsight has been working very well for me over the last year that I’ve been using it. It has some very useful and powerful tools built in that allowed me to get much more from my data. It’s also highly extensible. And the temptation of this extensibility finally led me to dig out an SSD from an old laptop to install Xubuntu Linux onto. PixInsight is well supported on Linux, Linux being the reference implementation, and there are new third party tools available for it on that platform. Now I have to live the dual-boot life.
The tool that triggered this was the recent release of NoiseXterminator, a neural network-based denoising algorithm. Reports from users on Cloudy Nights showed that the tool was indeed remarkable. Even though I try to collect a very large amount of data on a target, the light pollution in my area still leaves very noisy integrations that I must contend with. The recommended masked TGVDenoise and MultiscaleMedianTransform method works decently well, but it can lead to smudged details and mottled backgrounds. I really wanted to give this new denoise algorithm a try, and I have found it to be remarkable on the few test examples I have tried.
I went back to probably my toughest image yet, the Veil Supernova Remnant image taken with my widefield telescope, Nikon D5300, and a Baader UHC filter. With the poorly corrected bicolor stars and significant noise from limited integration, this image was a bear to process using the tools available in the older PixInsight base release. The first step this time around was to hit the color-calibrated integration with the NoiseXterminator tool. I am blown away by the results,
I find the denoising to be pleasant enough without totally demolishing detail or “plasticizing” the image. This really beats playing around with sliders in TGVDenoise and MultiscaleMedianTransform for an hour. My purchase of this plugin is solidified when the trial expires.
Starting off with a nice smooth image makes the rest of processing so much easier. I did have a bit of large scale mottling in the background after this denoise, but it was much tamer than I would usually get with my typical denoising scheme. I was able to kill it easily with some careful MMT thresholding. This gave me a nice, tight histogram peak that made stretching very easy.
Before stretching, I gave the next tool I was interested in a try. StarNet v2 is another neural networked-based tool that removes the stars from your image. And if I was going to test this tool on an image, this highly populated part of the Milky Way was as good an example as I had. It miraculously removed most every star, even the poorly corrected ones in the corners of the image.
I now had the starless nebula and another image containing the stars it removed. This made enhancing each extremely easy, compared to my old method of more slider guessing to build a star mask using the scale-based tool and a heavy application of artifact-inducing morphology. On the image only containing the stars, I reduced the size and brightness of the stars significantly using morphology and also desaturated them. I was then able to stretch each image independently, allowing me to really showcase all the detail that I could from the nebula while keeping the stars subdued and tight.
First, consider what I was able to do with the data on my last
futile patient attempt,
With the data now in a state unachievable to me previously, it didn’t take much more attention to finish the image to what I consider a pleasing result. A bit of dynamic range compression and saturation increase gave it that bit of “pop.” And the absence of the extreme field of false color stars lets you really focus on the target,