Using PixInsight TGVInPaint for star removal?

[Joel Shepherd]

Howdy - I've been playing around with using tone maps (a la http://www.arciereceleste.it/articoli/translations/75-narrowband-color-composition-eng ) for narrowband processing. Basically the idea is to remove the stars from the individual channels creating a "tone map" per channel, stretching them and doing an RGB combine, and then using the Ha channel enhanced by the others as a luminance layer. The idea, as I understand it, is to allow more, uh, spirited stretching and noise reduction on the weaker O-iii and S-ii channels without bloating the stars or introducing unwanted noise in the final result.

So, star removal. I've tried several different approaches, including MorphologicalTransformation, simply subtracting a star mask from the original, and using Straton. I haven't been very satisfied with any of them. Straton seems to work well visually, but it also seems to stretch the image a bit. The other two techniques seem to leave a lot of weird artifacts behind.

So now I'm playing with TGVInPaint which is part of a "development" module in PixInsight: http://pixinsight.com/forum/index.php?topic=8970.0 . I've made a bit of progress with it, but really don't know what I'm doing. Has anyone successfully used TGVInPaint for star removal, and if so is there a workflow or general process you'd be willing to share? (I've seen a couple video "tutorials" posted, but they are soundless and in any event I haven't been able to get the amazing one-click results they seem to show.)

[Joel Shepherd]

Answering my own question ... After dorking around with it for a few hours, I've been able to get some pretty good results by creating an *inverse* star mask and using that as the "support image", setting the Precondition check box and then slamming the "Noise reduction" slider all the way to the left (minimal noise reduction). With a decent star mask -- which is where the time investment is -- you can get almost one-click removal of stars from a linear image. Nebulosity is left intact. Sometimes a second pass with another star mask targeting very small stars is necessary, or use of the clone-stamp tool to dust up especially bloated stars.

Higher levels of noise reduction actually reduced the removal effect (and it seems to hit a limit at some point where repeated application has no effect), and increased the blotchiness of the result.

Pretty impressive once you figure out what to do: I'll try to post some before-and-after images tomorrow.

[wimvb]

I've never tried this module, but I'm looking forward to your results, and your tone mapped image.

[Joel Shepherd]

Here is the HA channel for Sharpless 2-86, a tone map made with repeated applications of MultiscaleLinearTransform (for nebula removal), star masks and MorphologicalTransform, and a tone map made with two applications of TGVInPaint with course and fine star masks and a very small amount of clone-stamping to clean up halos left by the 6-8 biggest stars.

Now that I've posted them side-by-side ... The first map, made with MorphologicalTransform is "cleaner": it doesn't have the speckling of small dim stars that are proving difficult to remove with TGVInPaint. I realize now I could clean up the halos on it, though they do stand out more than in the TGVInPaint. However, the first map also took a lot more effort: e.g., carefully clone-stamping away bits of nebula that leaked through into the star masks.

The second map, made with TGVInPaint, took much less effort but the speckling is very difficult to address. Or do further processing with. I suppose the next step is to try a combination of the two: using TGVInPaint to remove all but the smallest stars, and MorphologicalTransform for the smallest ones.

The processed image, using the non-TGVInPaint star maps is here:  http://www.astrobin.com/259374/ . I wish I'd recorded more processing details because I'm struggling to get back to a similar palette.

Anyway ... TGVInPaint, while quicker, may not give as workable a result which is a little disappointing. I may also play with it for star size reduction, which it seemed to do well in some earlier twiddling around.

Original HA stack:

Stars removed with MultiscaleLinearTransform and MorphologicalTransform:

Stars removed with TGVInPaint:

[wimvb]

I have found complete star removal to be impossible. Have just tried it for fun a few times, nothing I would use in any 'real' process.

The best attempt I've seen is this (scroll all the way down to ngc7000)

http://www.lightvortexastronomy.com/tutorial-reducing-star-sizes.html

I also get the artefacts that show in your image.

I've never tried very hard at complete removal, mainly because it doesn't suit my images.

Good luck

[ampleamp]

you might want to take a look at this thread which looks at the pros and cons of a few alternate star removal options