PixInsight Halo removal

[MattJenko]

I have been struggling with halos from images through my Oiii filter. This seems to be a common problem and while I could find no way of minimizing them mechanically, I am learning to live with them using PixInsight. This is a quick post to show a pretty effective technique I used in my last image to calm them down.

Below is an example. This is a calibrated and stacked Oiii image which has been stretched using a materialized STF autostretch (STF stretch applied via the histogram tool). It shows the halos in full effect throughout the image.

To address these, I process them at the Linear stage, and even before any linear processing. What is required is to use data from a different image to overwrite the halo. In this particular instance, I have a Green channel image which I found to be suitable for the purpose.

To match up the backgrounds, I took a copy of the Green channel data and used the tool LinearFit, with the Oiii image as the reference and applied the process with the default parameters to the Green image clone. This matches up the backgrounds quite neatly.

To create a mask for the Oiii image, I used an iterative approach using RangeSelection, with the Oiii image as the source. Lightness and Invert are highlighted and the Smoothness is set to a value large enough to spread out the areas of interest and small enough to still be seen. The first iteration looked like this :

I then repeat the range selection on each successive mask in order to get a good and accurate coverage of the halos. To help in this task, I found that there is a very useful trick in that you put one image directly over the other, and the use Ctlr-Pg Down and this flicks between the 2 images. This is extremely useful and allows you to get the mask to extend just beyond the halos themselves. Bear in mind at this stage, there will be areas of the mask which are over areas of the image which do not have halos. This gets addressed by using the CloneStamp tool, and with this tool, you simply get rid of the areas which do not have any halos, or where the halos are simply too close to the objects of interest in your image (in my case the Crescent itself).

When you are done, you can apply the mask to the Oiii image.

You end up with something like this :

We then are ready for a bit of image blending and to do this, I use PixelMath. I want to reduce the intensity where the halo is and not anywhere else, so I created the following expression : 

iif ($T - med($T) > (Crescent_G_clone - med(Crescent_G_clone)), ($T*0.6) + (Crescent_G_clone * 0.4), $T)

This basically states that if the difference between current pixel and the median value in the Oiii image is greater than the same ratio in the Green image (Crescent_G_clone in the expression above), then I know that the Oiii image contains a piece of halo. If that is the case, then I am replacing a percentage of the Oii image with a piece of the Green image. This means I can repeatedly execute the PixelMath expression over the Oiii image, and the halos are slowly reduced in appearance after each execution. The end result is below, and while there are outlines of the halos still present, this is much better and milder than before.

This image can now be used in whatever processing is going to be performed.