I thought it might be helpful to graphically illustrate how star removal works. My starting point was a small 100% crop from the original image above as it entered PhotoShop for the first time. I selected an area around the "Gulf of Mexico" and, as you can see, it has not been rotated to the usual orientation. Here is an animation of how star removal progresses:
The PhotoShop "Dust & Scratches" filter was applied successively with the settings 12/120, 10/100, 9/90, 8/80, 7/70, 6/60, 5/50, 4/40, 3/30, 2/20, 1/10, 4/8 and 2/3. I did a sanity check of the Preview at each pass but didn't feel the need to deviate from the 1:10 ratio until the last two passes which I felt were needed to get rid of the last stellar remnants. The final two steps shown in the animation were application of noise removal and some "de-purpling" using the Patch tool. I could certainly have spent more time on that last step but for today's purposes it was good enough.
Here is the initial frame and the one with the stars removed:
Subtracting one from the other (PhotoShop layers) resulted in a "Stars Only" version which I tweaked slightly to get the star brightnesses about the same as they were in the original:
It's not my purpose here to describe subsequent processing which was, in any event, quick and dirty but here are enhanced versions of the "just stars" and "no stars" images:
I took the opportunity to adjust the black levels of the "no stars" version and so hide some of the artefacts left over after use of the patch tool. I also tweaked the white balance of the "just stars" image slightly to remove a slight cyan tint. All processing was done entirely in PhotoShop.
I think it would be helpful to see the before and after versions as another animation as well as a still image of the end result:
Star removal has allowed me to retrieve more detail but there are costs. If you look in areas where fainter stars are in the nebulosity it seems pretty clear that in the final image they no longer appear as stars but instead as nebulosity! Whether that's good or bad is a matter of taste I suppose. If one believes that the objective is to produce a scientifically exact representation of the scene then it's a bad result, though how one defines "exact representation" when so many of the images we see are composites of RGB and narrowband data is open to question. On the other hand if one believes that the objective is primarily to produce a pretty result which doesn't deviate too far from what's actually up there then maybe the compromise is acceptable.
As I mentioned in the opening post I'm right at the bottom of the learning curve with this technique. My purpose here isn't to provide a masterclass, as I'm no master, but rather to show that star removal is actually not as difficult as it sounds and can offer benefits if applied judiciously.
P.S. On the remote chance that anyone would find it useful the initial crop is available as a PSD file here
. As a PSD, rather than a JPEG, it has the advantage of being 16 bit and in Adobe RGB so if you want to experiment and post your own results in this thread then feel free.
Corrected the white balance of the final "just stars" and dependent images.