Tarantula Nebula ReStarred

This is the third variation on the Tarantula generated from the abundant 40 hrs of HDR exposures described in https://astrob.in/kj33x7/0/. In fact, this is the same image as in that reference, except I have re-inserted the stars down to 16th magnitude.

In this discussion, I want to describe some of the issues that stars cause in images of nebulae. The first one has to do with the advantage of removing the stars before enhancing the nebulae; the second is about the problems that arise when you try to remove the stars (using StarNet); the third is about the very substantial problems encountered when you try to re-insert the stars in the processed nebulae-only image. These are large topics but I will give them only a light brush, so don’t despair.

Maybe it’s just my eyes, but stars seem to hide or subdue the details in a nebula…relegating the nebula to a background, less prominent role in the image. Being that the nebula was likely the reason for the image in the first place, its diminished prominence seems counter-productive. Compare this image to the one on the Astrobin page referenced above. To my eye, there are vastly more subtle complexities in the faintest reaches of the nebula in the starless image than in this one. But even the current image, with replaced stars, shows more nebula detail than an image processed without first removing the stars. (The alternative to removing the stars—masking them—has different strengths and weaknesses. But overall I feel I get better results and fewer artifacts by removing the stars than simply masking them.) To be honest about using StarNet, if you have a small refractor, life is good. If you have a reflector with a secondary support, then not so much. You will have difficulty removing the star spikes. There is (was?) an effort to retrain StarNet on reflector data (https://stargazerslounge.com/topic/357814-starnet-training-for-reflectors-lets-collaborate/) but it may have stalled?

The principal advantage of removing stars before doing much processing is that adjustments to nebulae colors, contrasts, and intensities are not constrained by the colors and dynamic range of the stars.

Moving on to replacing the stars in an image, I see some misunderstandings that are cause for concern. Some imagers seem to assume

(Original image) – (Starless image) = Stars

It does not. The starless image has had the star locations patched—changed in color and brightness. Therefore, in the above equation, the subtraction is the difference between the original image and the color/intensity altered image, which does not yield the stars as they appear in the original image! And it gets worse. The Starless image inevitably gets enhanced. Its colors, saturations, intensities are altered. Now, you cannot add the stars back to (Enhanced Starless image) and get the original star colors or intensities. As far as I can ascertain or imagine, the only way to get the correct star colors back into the enhanced image is to copy the stars from the Original image after erasing the patched area in the enhanced image. That’s what has been done in the image posted here. There are many rocks in this path too, and I’m still trying to push them off the trail. For instance, do you want/need to have a PSF for the copy-back template? How do you best/rigorously match re-inserted star intensities to the Altered Starless image? What about proper star sizes? And so on!

Folks, I’m open to suggestions and would be happy to send the PixelMath icon I’m currently using to anyone who wants it. (Send me a private email through Astrobin to request it.)