Pixinsight HSV Separation Pink Star Cores

[Scooot]

Its been very frustrating not being able to use masked stretch because it always creates pink star cores.

I've attempted to correct them with PixInsite's Repaired HSV Separation script and Channel Combination before the Masked stretch but whatever settings I use I still get pink star cores. So I'm wondering whether my version of Pixinsight is faulty or I've got a wrong setting or it just doesnt work with my data for some reason.

I wonder if someone else wouldn't mind attempting it. Here is a link to a linear crop of Aldebaran. The image was taken with my Canon 450 and Samyang 135mm F2 at F4. I did post a similar thread on the PixInsite forum but didn't resolve it so thought I'd try here.  

https://www.dropbox.com/s/mmr10f9qi2y1g98/Aldeberan Crop Linear.xisf?dl=0

[glowingturnip]

I had a quick look - the problem is that the raw file isn't colour-balanced - you can see it very easily if you turn the screen stretch off, you only see the core of Aldebaran and it's unmistakeably magenta.

Try something like this - the colour calibration first, then the masked stretch.  Play around with the parameters, setting the white reference upper limit too high seems to turn it green.

[Scooot]

Thanks Stuart.

 I had colour balanced the image using photometric colour calibration prior to taking the crop. The colour balance of the overall image looks quite good on an Stf stretch and if I process the image without using the masked stretch it also seems quite good. However maybe the colour balance routine didn’t do Aldebaran correctly because it was partly saturated? I think it’s showing as magenta in the raw file because it’s over exposed in the centre.

[alan4908]

I'd suggest you try the Repaired HSV separation with the default values but the pixels set to 15.  This will give you an orange star when you view the resultant image (unstretched).

In terms of stretching, I personally never use masked stretch since I don't like the result.  Rather I would suggest using arcsinhstretch for the RGB and combining it with a synthetic Lum using LRGB combination. 

Alan

[Scooot]

6 minutes ago, alan4908 said:

I'd suggest you try the Repaired HSV separation with the default values but the pixels set to 15.  This will give you an orange star when you view the resultant image (unstretched).

In terms of stretching, I personally never use masked stretch since I don't like the result.  Rather I would suggest using arcsinhstretch for the RGB and combining it with a synthetic Lum using LRGB combination. 

Alan

Thanks very much Alan, I’ll try that.

I don’t have arcsinhstretch in my version of PI for some reason. I think I’m going to have to uninstall and reinstall it but I’ve been putting off doing it.

 

[Scooot]

14 hours ago, alan4908 said:

In terms of stretching, I personally never use masked stretch since I don't like the result.  Rather I would suggest using arcsinhstretch for the RGB and combining it with a synthetic Lum using LRGB combination. 

Alan

Hi Alan,

I didn't need to reinstall Pixinsight for arcsinhstretch, I went to process install modules. It wasn't listed but I used it's search and it found it. So it had been downloaded at some time but didn't automatically install. I'll try it out now. Thanks for the suggestion.

[IanL]

MaskedStretch is useful to retain colour in stars as a single global stretch tends to oveestretch them to the point where the cores are (nearly) white and no amount of further processing will get the colour back. One downside is that the rest of the image lacks contrast so you have to do more work to boost contrast, etc.

You can try masked stretch and a normal stretch on clones of the image and then use a star mask to blend the two, but it can be tricky to get a mask sized appropriately for large, medium, small and tiny stars. It might work better to do a masked stretch and then use a star mask to selectively apply further stretches or curves just to the background.

The unwanted colour tints seem to be a byproduct of the fact that the stars are different diameters in each of the RGB channels, especially in DSLR/OSC images where you can't adjust the exposure time per filter to balance the sizes (if you don't you can also get colour casts when combining your LRGB image when using mono and filters). When the masking process is operating each channel gets a series of stretches that accentuate the problem as each channel ends up with different star profiles, leading to the magenta cast. Usually it is the medium stars that suffer most in my experience.

It's pretty hard to avoid and harder to get rid of once created. I've tried things like extracting the colour channels, applying  different star masks and using morphological transformation to shrink the stars so they match sizes across channels, but it is not ideal as clearly tge star has a Gaussian or similar profile.

You can also try selective curves and a mask to desaturate the magenta, but this risks unnatural star colours. There is no magic bullet as the best approach seems to vary from image to image.

[Scooot]

Thanks for the reply Ian.

I agree its not easy to deal with the pink stars once they've been created which was the attraction of the HSV Repair tool, if only it worked  

[wimvb]

14 minutes ago, Scooot said:

Thanks for the reply Ian.

I agree its not easy to deal with the pink stars once they've been created which was the attraction of the HSV Repair tool, if only it worked  

It worked for me, afaIk. I used default settings, but radius set to 32. Aldebaran is a red giant, so it still comes out red, but not pink.

The only problem I had was getting a proper colour calibration with this tight crop. Standard colour calibration made the star too blue.

[Scooot]

2 hours ago, wimvb said:

It worked for me, afaIk. I used default settings, but radius set to 32. Aldebaran is a red giant, so it still comes out red, but not pink.

The only problem I had was getting a proper colour calibration with this tight crop. Standard colour calibration made the star too blue.

Thanks Wim, it came out red for me as well but turned pink with the masked stretch, or are you saying it was red after the masked stretch?

[wimvb]

4 hours ago, Scooot said:

Thanks Wim, it came out red for me as well but turned pink with the masked stretch, or are you saying it was red after the masked stretch?

Much less pink, but not completely red either. But remember: red + brightness = pink. A star as bright as this will need to be rendered white. Keeping colour in the core won't be convincing.

Comparison of Masked stretch + curves transformation and repeated use of curves transformation.

Using only masked stretch leaves a false coloured core, even after HSV repair. But this is minimised in curves transformation.

 

[Scooot]

38 minutes ago, wimvb said:

Using only masked stretch leaves a false coloured core, even after HSV repair. But this is minimised in curves transformation.

Ah,this is what I was finding. I think I’ve been expecting too much from the HSV repair  

I’ve been using mostly a combination of Histogram Transformation and exponential transformation although recently ive found just copying an adjusted STF is a good starting point. 

Thanks very much for the work you put into explaining, it’s much appreciated. This was my finished effort on this image. I certainly didn’t want a coloured star core but I did want plenty of colour in the main stars to highlight the formation. Still learning, it’s good fun