Removing stars using the starnet++ program

[STEVE333]

I've had some very nice success using the new starnet++ program to remove stars from a stretched image. The following has worked for me"

I downloaded Starnet++ from the link below which was provided by Geethq.

https://sourceforge.net/projects/starnet/

I found the following works for Windows computers:

1) After downloading, the starnet++ files will be located in a folder named starnet. Don't forget to extract the files.
2) The program only works on 16 bit unsigned .tif files.
3) Save the file you wish to remove the stars from into the starnet folder (for example, FlamingStar.tif) in 16 bit unsigned format.
4) Open the Command Prompt.
- I typed capital C in the search box in the lower left corner of the Windows 10 home screen.
- In the popup window click on Command Prompt.
- In Command Prompt type cd\the path to the starnet folder\starnet (e.g., cd\Documents\starnet ) then hit enter.
- Type starnet++ OriginalFileName.tif StarlessFileName.tif where OriginalFileName is the name of the file you want to process.
EXAMPLE: starnet++ myimage.tif mystarlessimage.tif (notice that there are spaces between starnet++, myimage.tif, and mystarlessimage.tif.
- Once the filenames are typed in hit ENTER and wait till the processing is completer. Progress is shown as %done.

When the Command Prompt says Done, the resulting starless image will be located in the starnet folder.

Attached are examples showing an original image and the starless version produced by starnet++.

Using PixInsight, I typically do sharpening (MLT), contrast enhancement (LHE), color saturation boosting with CurvesTransformation, and noise reduction (TGV) on the starless image. Your processing software and enhancement choices may be different.

Once the starless image is "enhanced" I use PixelMath to combine the starless and original image with the simple expression

max(original, starless)

Hope this makes sense.

Steve

 

[msmythers]

Steve

See this posting.

https://pixinsight.com/forum/index.php?topic=13691.0


Mike

 

[STEVE333]

Thanks Mike - Appreciate the input.

I saw the thread during my research. Since there still seemed to be some issues with the PI instance, I decided to share the "older" standalone approach that should work for most Windows users.

I've really enjoyed using this program with my new "dual narrowband" images that are quite similar to HOO data. Sometimes the starnet++ program will degrade some of the fine detail, and, sometimes it doesn't. Just depends upon the particular target. When it works it's amazingly simple to use.

Steve

 

[Lammertus]

Hi all,
I have been using the commandline IF for some time now and have seen this thread, now it is installed in Pixinsight as per the readme.txt file. Some strange behaviour though when you stretch an image with Arcsinh, then there are nearly no stars detected by starnet++ where if using the Histogram stretching or STF -> Histogram and applying, all seems to work well! Just thought to share this with you.
Regards,
Mert

 

[dave_galera]

Hi Mert,

Suggest you add this post to this thread where it can be more easily picked up by Nikita.

https://pixinsight.com/forum/index.php?topic=13691.msg83068#msg83068

 

[Lammertus]

Thanks Dave, I just reposted on the other thread!
Regards,
Mert

 

[STEVE333]

Just an update. I've installed Starnet++ into PI so that it shows up as one of the available Processes. Much easier to use, and, so far, it has accepted all images formats (including .xisf 32 bit). So, no longer necessary to save a 16 bit .TIF image in the starnet folder and then run Starnet++ from the Command Prompt window. Also, you keep all the precision of the original image.

Just FYI

Steve

 

[ghethco]

I have been finding that StarNet does not remove all of the stars in my images. See below.

There aren't really many dials to tweak with StarNet, unless I'm missing something.

Has anyone seen a similar result and found a way to deal with it? If I don't do anything, I end up with artifacts after the stretching and the stars being added back in.

Gary

 

[JPetruzzi]

Starnet is for use on non-linear images, it works really poorly if you try to use it on something before stretching

 

[ghethco]

Yes, I'm aware of that. This image was fully stretched before using StarNet.

One thing is that the stars that remain are mostly toward the edges of the field. My field flattener doesn't work 100%, so can't help but wonder if it is the blurring/distortion toward the edges of the field that is causing this. Doesn't really help to know that though. The alternative is just to not use StarNet.

 

[FlapAstro (aka rakla)]

I had less than optimal results (similar to yours) when using MaskedStretch. Not sure that is your case, but if so you might want to try the "classic" way and see if that makes a difference.
/Ralf

 

[JPetruzzi]

Yes, I'm aware of that. This image was fully stretched before using StarNet.

One thing is that the stars that remain are mostly toward the edges of the field. My field flattener doesn't work 100%, so can't help but wonder if it is the blurring/distortion toward the edges of the field that is causing this. Doesn't really help to know that though. The alternative is just to not use StarNet.

Ah, sorry--I misunderstood because you had said "...end up with artifacts after the stretching". Also, the boldface type was accidental.

 

[ghethco]

@JPetruzzi - No worries, I see how you got that idea.

@ralf -- the workflow I'm following is from this video. DBE, channel combination, then ATF stretch permanently, then some minor color correction, then extract the stars using StarNet. I haven't tried "MaskedStretch". By "the classic way" did you just mean using CurvesTransformation? That's what is used in the video, not MaskedStretch.

Maybe I should have described in more detail what happens at the end. All goes well until I add back the stars, at which point the ones which weren't removed have little dark haloes around them. I can post an image of this if we get any further. The image above is right after star removal.

Gary

 

[FlapAstro (aka rakla)]

Gary, yes that´s what I meant by classic STF + HT.

I typically do something like DBE / ABE, PCC, (optional Deconvolution), linear noise reduction using MLT, SCNR, followed by HT to obtain a non-linear image.

I then use Starnet before any other non-linear operations, such as TGV, CT, etc..

Mind you, I use an OSC and not a mono camera.

/Ralf

 

[rroberts]

I certainly see the same artifacts with star halos making it into the starless image. I'm not sure how much I'm going to object to that when I'm done, but I expect I'm going to have some problems. And I happen to be working on the same field as you right now, NGC 7000

Here is a close up around Deneb. The left is the "starless" image after I've done some processing that involved a bit of masked MLT noise reduction, masked curves transformation on the saturation and another pass to reduce some mid-scale chrominance noise with MLT. The image I used for starnet++ similarly already had some masked MLT noise reduction on the first three layers.

The main thing that I noticed immediately is what looks like stride/block boundary artifacts. While I can easily believe that adding the initially split images back together would result in the artifacts canceling out, I'm pretty sure that adding them back after doing processing separately won't.

I should add, the masked processing on the left prevented any blur/noise reduction, but it certainly ended up included in the saturation enhancement, so that boundary is definitely going to stick out when I put them back together.

Anyone try StarXTerminator? I heard Adam Block make a passing remark about using it in one of his videos....

 

[rroberts]

I should have also mentioned that I did to the initial pre-Starnet++ stretch the "classic" way, which was an unmasked stretch via HistogramTranform. The MLT noise reduction was done before that, but pretty minimal. It was done using a non-linear stretched luminance mask, but since that was noise reduction only, I don't think that should matter.

 

[pfile]

SXT is pretty good. and runs on the GPU on all platforms, i think. but you need to be running at least Big Sur on the mac.

 

[rroberts]

Well, it seems to do better. I just downloaded the trial to, well, give it a trial. I'm running on an Ubuntu VM with an old GTX 1060 passed through.

First the full image comparison, starless from Starnet++

vs StarXTerminator

Deneb is still a problem, but there are several smaller stars that no longer show up. At this scale, it's a little hard to tell, but even here you can see the yellowish blog of Xi Cygnii (left of "Mexico") that shows up in the Starnet++ image but is nearly invisibl in the StarXTerminator. There is another small dot of a star between the Pelican "shoulder" and the image boundary that also disappears.

You will notice that the left side of the image is much more mottled in the StarXTerminator than in the Starnet++ image. At first I though that was going to be a problem, but looking at the preview panes, I think StarXTerminator is doing okay; it's smooth enough there than I'm not going to have weird artifacts and I think the algorithms just disagree about how much of the brightness is star halo vs nebulosity.

Also, even at this scale, you can that weird sharp boundary on Deneb is gone. I think I'll finish working on this image before I decide, but I'm probably going to open my wallet for this one.

 

[rroberts]

Here are the preview detail pains for comparison. I think this region will be fine and probably look better with StarXTerminator. It probably is star halo in this region, but I'm not a pedant for "correctness" here....