Attempting to use MT for Red Star 'Halo' reduction

[johndias]

I am using a new filter (Astronomiks CLS-CCD clip in with my Canon 60Da) and my images are showing a red halo.  Basically, the red channel seems to be creating much bigger stars.  I followed some suggestions to extract the red channel, create a star mask and use MT to reduce the star size but even though I have run MT at 10 iterations (using Erosion) many times, the reduction is insignificant.  Even though i think I am getting close, when I recombine the RGB the red halos are still very prominent.  Any suggestions?  Screenshot of my workspace, showing the main image with the halos, the R and G channels, and the last MT settings I was attempting.

[johndias]

bumping.

If anyone can provide more detail on the MorphologicalTransform process or a better method for reducing stars, I would be very grateful! 

[avastro]

Hi,
Post a link to the Fits Red image, to allow people to download your image and do some tests and probably propose you a solution.
Best
Antoine

[johndias]

Thank you for the suggestion.  Here is a link to the color fits file.  My workflow so far:

Channel Extraction
Star Mask on R channel
Apply mask and run MT with Erosion, circular structure (I've tried different sizes, not sure what that means) and 10 iterations. 
Channel Combination (assuming I get the stars small enough in R)

I do see a very gradual reduction each time I run MT this but I am hoping there is a way to make a more aggressive reduction.  I followed Harry's video (Star Mask 1) to get the basics down.

By the way, if this is a lost cause due to the R being so far out of focus I am willing to accept that.  I realize that some things just cannot be fixed.

[avastro]

Hi John,
I've a look to your image, your image is lacking a good focus, mainly in the red channel, that could be related to the Astronomiks CLS-CCD clip, or the quality of your optics who focus differently depending of the wavelength. As a side effect the stars alignment is not as good expected. I recommend you also to integrate more images to improve the ratio signal to noise.
I worked in a crop of your image, extracting the channels and reducing the stars size in the Red channel.
The first step is to do a StarMask covering perfectly the stars to be reduced, with the mask showing these stars and protecting all the remaining then using MT with Morphological selection instead of erosion, it's more efficient in my opinion, apply MT with 2 or 3 iterations. Look at the forum about MT selection for more information or links like this one are invaluable source of information.  http://pixinsight.com.ar/en/
Then combine back the channels using the Red modified.
I reduced  the noise using TGV denoise and stretched with histogramTransform.
Here is my try.
Best

[johndias]

Antoine, Thank you! 

I was close to the solution but I did not understand the purpose of the Amount slider.  I thought that a lower value would provide more reduction, but it seems to be the opposite.

Yes, my capture isn't ideal.  But thanks to you I can salvage something from it and learn along the way.

[avastro]

Hi John,

MT with Morphological selection can be used in two way, with a selection value inferior to 0.50 that will decrease the star size (erosion), over 0.50 will increase the stars (dilatation), amount allow to adjust the value applied.
In the previous snap, selection 0.38, do a rather important erosion, amount 0.7 is 70% applied; structuring element is circular with a 5 by 5 elements, you can test a structuring element 7 by 7 for bigger stars, also use a different geometry form as diamond structure or combined using 2 ways, depending of the star to be corrected, also using manage, save your structures, pick new one ...
Worth to invest some time to learn better this useful tool! 
Best

[johndias]

MT is working very well now that I understand, thanks.  The problem I am having is with getting my star mask to pick up the smaller stars.

[avastro]

Here is an StarMask example with your linear image. Statistics process, note the mean value, open StarMask process, Copy mean value in noise threshold and mask processing midtone,adjust mask generation smoothness around 8 depending your image. done!
For difficult case with an RGB image: extract the CIEL L component, apply a HDR Multiscale transform 3 or 4 layers 2 or 3 iterations to it, then the same process as above  to the L compment to generate the starMask.
PI allow many way to generate a mask with different tools, do a search in the forum you' ll find a lot of innovative techniques.
Best
Antoine