Re-insert stars in starless image

[joelshort]

I could use some advice.  I'm working on an image that unfortunately has some severe, sharp, arcing reflections.  I had the idea to use StarNet++ to create a starless image so that I could more easily clone out those reflections, and then reinsert the stars back into the starless image.  However getting the stars back in the image is proving harder than I thought.  The main problem is that the very bright stars in the image are proving very difficult to include so that there isn't some sort of halo around them, and creating an accurate star mask is exceedingly difficult to produce. 

Any ideas on how to accurately reinsert those stars in the starless image?

Attached is a screenshot of the images I'm working with.

[dave_galera]

Don't use star mask, use pixel math as below:

To save the stars as a separate image use PixelMath Original - Starless = Stars

To put stars back in again use PixelMath Starless + Stars = new image with stars

The Stars image is in effect the best star mask you will ever get

[joelshort]

Don't use star mask, use pixel math as below:

To save the stars as a separate image use PixelMath Original - Starless = Stars

To put stars back in again use PixelMath Starless + Stars = new image with stars

The Stars image is in effect the best star mask you will ever get

There's two problems with this approach.  Simply subtracting the starless image from the original image still leaves those large arc reflections which is what I'm trying to remove in the first place.
 The second problem is that the large stars in the image are not being fully removed by StarNet++ so simply subtracting the starless from the original is problematic too because it subtracts real star data from the original image. 

[dave_galera]

Yep that is a problem with StarNet++ it  doesn't like very large stars.

[wadeh237]

Have you considered doing a histogram adjustment so that the image with stars has slightly less brightness than the starless image, and then doing PixelMath to keep the higher pixel values between the two images?

That should get the stars from the first image, but only in places where they are brighter than the starless image.

[joelshort]

Have you considered doing a histogram adjustment so that the image with stars has slightly less brightness than the starless image, and then doing PixelMath to keep the higher pixel values between the two images?

That should get the stars from the first image, but only in places where they are brighter than the starless image.

Yes, that would work, but again it does not eliminate the arc reflections in the original image.  PixelMath takes that bright parts of the original image including the arcs into the new image.

[jtalbot]

Hi Joel,
This seems like a complicated problem.   Just a thought.  One thing that may work but take a bit of work is to create your stars image which also has the arc in it.  Make a copy and binarize that to the point where the fainter stars around the arc disappear. At that point you could clone out the rest of the stars (make the rest of the image black).  Now you have a mask that just includes the arc.  Maybe subtract this from the stars image to get an image of just the stars without the arc which you could add back into your starless image.  It may take a bit of finessing the mask to get it to work perfectly since I see thee are stars within the arc.

Jon

[joelshort]

Hi Joel,
This seems like a complicated problem.   Just a thought.  One thing that may work but take a bit of work is to create your stars image which also has the arc in it.  Make a copy and binarize that to the point where the fainter stars around the arc disappear. At that point you could clone out the rest of the stars (make the rest of the image black).  Now you have a mask that just includes the arc.  Maybe subtract this from the stars image to get an image of just the stars without the arc which you could add back into your starless image.  It may take a bit of finessing the mask to get it to work perfectly since I see thee are stars within the arc.

Jon

Thanks for the idea Jon.  I gave up and just cropped the image. 

[wadeh237]

Have you considered doing a histogram adjustment so that the image with stars has slightly less brightness than the starless image, and then doing PixelMath to keep the higher pixel values between the two images?

That should get the stars from the first image, but only in places where they are brighter than the starless image.

Yes, that would work, but again it does not eliminate the arc reflections in the original image.  PixelMath takes that bright parts of the original image including the arcs into the new image.

Sorry, I misunderstood.  I thought that you separated the stars so that you could deal with the arcs on a starless image, and that you were now trying to bring the stars back...

[STEVE333]

I also have had problems with large bright stars. One thing that has sometimes worked is:

• Make a clone of your image
• Run Starnet++ on clone to create the starless image
• Use CloneStamp to cover the halos around large bright stars in starless image
• Process starless image (sharpen, etc.)
• Use max(starless, original)

This has often worked for me.

Steve