When to remove stars during process?

With Starnet2, there's now the ability to remove stars while the data is still linear. Given this new feature, when is the best time in the workflow to remove stars and are there any tradeoffs? Thinking of this in context of my SHO workflow.

It was there already, as a script, so it isn't really a new feature. This said, I'd warrant that the answer very much depends on what you want to do with it. I wouldn't do anything until after flattening and colour calibration. Further cosmetic correction would have to be carried out best in the linear image. Same for noise reduction and deconvolution.

I find that it depends on the target. Not one size fits all as sometimes Starnet does not always do a good job at removing the majority of the stars. Also depending on your LP conditions I find removing the stars introduced more noise than creating a mask.  It could be I have only been doing this for 1.5 years but always trying to refine my post processing skills.

andrea tasselli:

Andy Wray:
I'm not sure I quite got your response. I guess the question was whether to do star extraction in linear or non-linear. I tend to stretch the image until I get stars roughly like I want them. I then extract them so that I can work on the starless image so that I can extract the nebula detail. I then add the stars back in using pixelmath. I may have modified the stars to either reduce them or soften them prior to adding them back in.

The answer is/was: it depends. Some best done in the linear phase, some in the stretched phase.

Thanks for the answer, but maybe a more direct question would help:  do you ever personally create a starless image in the linear phase of your processing using Starnet or starexterminator or do you tend to stretch the image first a bit into non-linear territory?

Andy Wray:
Thanks for the answer, but maybe a more direct question would help: do you ever personally create a starless image in the linear phase of your processing using Starnet or starexterminator or do you tend to stretch the image first a bit into non-linear territory?

I do. Whenever I had to perform deep cosmetics to the image (e.g., removal of cosmic ray hits and/or residual blemishes from imperfect flats, hot pixels). Whenever I have to create a star mask for deconvolution and run deconvolution itself. And denoise using MLT. All in the linear phase. Then for other things after the final stretched image is created (e.g., additional colour balancing in the stars, unsharp mask of nebulosity and so on and so forth).

When I process SHO images I first use auto-stretch in Screen Transfer/Histogram then extract the stars with StarNet2, then ABE/DBE/TGVDenoise/LRGB Combine…

Any image manipulation software that only gets your image, without any other info, will work more reliably on a linear or lightly stretched data. The software must make an assumption, that the data is reasonably linear. The problem is there are an infinite number of ways to treat your data non-linearly, so including that option in a program is a non-starter. The exception is if the program knows exactly how your image was stretched, which is only possible if the program is also doing the stretching.

That being said, if you use a tool like this early in your processing, any problems will be magnified by any non-linear treatments later in your processing. And let's face it near the end of your processing, nothing is likely to be linear.

I find with any of these additional processing algorithms that the best place to work is after your primary stretches, masking, etc., once most of the heavy lifting is done. For one you can see immediately the effect of the star removal tool on your final image, and fine tune it. If the tool messes with your image early on, the problems only get amplified and are impossible to correct later on. We have all been in that situation, we mess up and have to go back to the beginning–as it is very difficult to correct problems introduced early in the processing. 

My experience is to get the image 90+% there. Further, you can also try to prepare you image for star removal. Reduce star size and halos as much as you can first, all of the star removal tools have trouble with halos and big bright stars. Making stars smaller helps too, apply those tweaks first.  Do anything you can in processing to make the star removal easier. You will still have to do some manual tidying up, but much easier to do this as a tweak on the mostly final image.

AI tools are wonderful, but we are smarter, and can make the AI's job, and ultimately our work easier if we do our due diligence. Do most of the work yourself and you will be amazed at how well these AI tools do their job…
Rick