Starnet2 vs. StarXterminator

Using tools like Starnet and StarXterminator have revolutionized image processing in the last few years. They make it simple to extract stars, process the rest of the image and then restore the stars making it possible to do things that just weren’t possible when the stars within the image. Or if possible, at least not nearly as easy to accomplish.

But, like any tool, they aren’t perfect. The original version of Starnet, Starnet++ was slow and often left artifacts that made it unusable for many images.

Then StarXterminator arrived and it was both faster and produced generally better results than Starnet++. However, now Starnet2 is on the scene so how does it compare with StarXterminator?

The first, obvious difference is that Starnet2 is a free add-on to PixInsight (there also is a standalone command line version) while StarXterminator is a commercial product that costs $60 for either a PixInsight plugin or a Photoshop plugin. In this comparison we’ll be examining the PixInsight plugin versions.

To install Starnet2 is currently a manual process. It’s not hard but it is a bit cumbersome. The author says that it will eventually be integrated into PixInsight itself as the original Starnet++ was but there is (so far as I know) no timeframe on when that will happen.

StarXterminator uses PI’s repository system so once you tell PixInsight about the repository it will keep you up to date. Nice simple and reliable.

Once you have them installed they operate very similarly. Apply the process to an image and it will remove the stars. Either can optionally generate an image with the extracted stars and either can optionally work on linear images. Starnet2 does this by internally stretching the image and then “unstretching” it once complete. I don’t know whether StarXterminator works this way or operates directly on the linear data.

In the video we only look at non-linear images In theory there might be advantages to working with linear data but since both are capable of leaving star related artifacts behind you can blend the stars back in more easily if both the image and stars result from the same stretch. I have the scars to prove this one. It’s possible this may not be true in the future but it is true for now.

In the video we look at three different telescope/camera combinations:

  • Ceravolo 300mm Astrograph with FLI ML 16200
  • Planewave CDK14 with FLI PL 16803
  • Takahashi TOA-130 with QHY268M

The first two are large aperture telescopes with central obstructions and CCD cameras while the last is a refractor with a CMOS camera.

In all cases the images were minimally processed before removing stars. For the broadband data this was:

  • crop
  • RGB combination
  • deconvolution (only all the Ceravolo images and the NB for the Planewave)
  • dynamic background extraction
  • stretch

For the h-alpha images it was:

  • crop
  • deconvolution
  • stretch

In most cases that stretch was via the Histogram Transformation process but more recently I’ve begun using the Generalized Hyperbolic Stretch script so the M100 image was stretched with that.

Generally there are two types of problems that can result that can be broadly classified as false positives and false negatives. The case positive case happens with the star extractor improperly recognizes something as a star and removes it. This is most likely to happen when imaging fields with small galaxies in them. The good news is that these extracted galaxies are likely to end up in the extracted stars image (if you created one) so they aren’t lost.

The false negative case occurs when a star or portion of a star is left in the starless image. This one can be more problematic as it can make the starless image tough to process. It can lead to a lot of clone stamp “fun” or reverting to different methods that leave the stars in place. However, both products are getting better at leaving more manageable artifacts behind. By “manageable” I mean that the “blog” is smaller and generally matches the background and texture of what surrounds it. However they aren’t perfect.

I tested this in PixInsight 1.8.9 on an iMac Pro running MacOS 12.3. The iMac Pro has a 10 core Xeon running at 3GHz and 64GB RAM. The PI temp space is on a dedicated SSD. The version of Starnet2 is, well, Starnet2. The version of StarXterminator is 1.2 using AI 8.

Starnet2 appears to be making some use. of multiple CPU cores but not the GPU while StarXterminator appears to be single threaded but does use the GPU. Coincidentally both appear to take about the same amount of time to process the same image on my system. This varies from around 46 seconds to 90 seconds depending on the image size.

Here are some screenshots showing the results. First up is an image of M100 from the Ceravolo. Here, Starnet2 does the better job. Both leave artifacts but the Starnet artifacts are less objectionable than the StarXterminator artifacts. Starnet2 did remove some small galaxies that is should have left in but they did show up in the extracted stars.

In the image above Starnet removed the galaxy circles in pink. StarXterminator left behind artifacts from extracted stars (circled in yellow).

In the two images above,, StarXterminator is on the left and Starnet2 is on the right. StarXterminator leaves a bit more of the star behind in this hydrogen alpha image. However it is not as problematic as it was in the broadband image.

In the LRGB image from the Planewave CDK14 both star removers struggled though Starnet did a slightly better job. StarXterminator left a star in and left brighter halo remnants. Neither did especially well on this image. This image is a bit of a torture test. It has a lot of stars and has very bright stars.

I only included one image here because they were virtually identical. Both extractors did a great job. I might give a slight edge to Starnet2 but it is very, very slight.

This is a bit of an eye test but this is from the Takahashi/QHY268M combo. Starnet2 is on the left and StarXterminator is on the right. The difference is subtle but StarXterminator leave small star shaped circles in the image. This image has a lot of small stars and Starnet2 handles them somewhat better. However, if you plan to put the stars back then the extracted stars should “cover up” the circles and it won’t be a real world problem. However if you plan to leave the image starless then Starnet2 is the better choice.

This is another eye test. Starnet2 is on the left and StarXterminator is on the right again. In this case Starnet left a star blog that StarXterminator didn’t.

What conclusions can we draw from this?

Deconvolution may make things harder for StarXterminator — or at least the way I deconvolve may make things harder.

With that out of the way, neither is perfect. On broadband data Starnet generally did a better job with the exception of removing some small galaxies that it should have left. Fortunately they did end up in the extracted stars so they weren’t lost. StarXterminator left more and more noticeable artifacts in all of the broadband images looked at.

For narrowband images the results were much, much closer. Both did a great job , nearly perfect in all cases. There were slight differences but both were terrific.

I think it’s safe to say that the fewer bright stars you have in an image the better both will do but currently, Starnet2 handles them better than StarXterminator. Starnet2 seems to handle central obstructions and diffraction spikes better than StarXterminator, at least for broadband.

Both are good tools. Which will run faster for you will depend a great deal on your hardware so I can’t generalize except to say that StarXterminator will use your GPU and Starnet2 won’t without you finding a specific library that actually uses the GPU and replacing it in your install. Starnet2 appears to use multiple cores for processing but that’s empirical evidence based on watching system metrics while it runs. I don’t know what it’s doing with those cores. Starnet2 is definitely faster and much, much better than Starnet++. The difference between Starnet2 and StarXterminator is smaller and will largely depend on your data.

I bought my own copy of StarXterminator and the author of neither product had any input (or even knowledge of) this review, however, I’d be happy to supply problematic images to either author to improve things for future updates.

I think Starnet2 is best for most people. The price is right and it usually does the better job, at least for me. However, you do need to be a bit more computer literate to install it. Hopefully that will change in the future. StarXterminator is a commercial product but I think it provides good value and it seems to be well supported by its author. It never hurts to have multiple tools and to be able to pick the one that works best in your situation.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s