PI Deconvolution Question

[peter_4059]

I've been playing wit PixInsight Deconvolution today but have been unable to get a result I'm completely happy with. This issue is the medium size stars get a white dot in the centre and this gets worse with more iterations. Is there a way to prevent this effect in Decon settings or is there a post-deconvolution fix?

Cheers,

Peter

[RickS]

Are you using a deringing mask, Peter?

[peter_4059]

Yes I'm using a deringing mask created with the star mask process and a lum mask using an STF->HT stretched clone of the original

[mostschaedel]

Let us have on dropbox your fits file you want to deconvolute.
Will try to solve the issue and provide the parameter...

The finding of the correct parameter is very tricky.

Gerald

[RickS]

Hi Peter,

I usually see some brightening of star cores from decon but nothing that looks bad, so I assume this is something different.

If, as Gerald suggested, you're willing to put a copy of the image somewhere I'd be happy to have a play with it.

Just a thought... is the data linear or stretched?

Cheers,
Rick.

[peter_4059]

Thanks for the offers. The image is still linear. I've been sick today but will get it on dropbox when I get a chance.

[codemonkey]

I get a hard edge around saturated cores, which might be what you're talking about, when I don't protect them during decon.

When applying decon I create a luminance mask like you do, although I make the background near black, and try to push the midtones a lot so that I get good decon on my target and less "noise enhancement".

After creating that mask, I then create a star mask (from the original) to capture any of those stars with saturated cores (I usually go with a scale of 5 for this). I then subtract that star mask from the luminance mask I created initially, and then use the resulting mask to protect the image when performing the decon.

[peter_4059]

I've put the files in dropbox:
https://www.dropbox.com/sh/x8z6q2c9kpq8ibo/AAAk6kbMljItMb0_6nBmA7Rta?dl=0

I've included:
1. the image pre-decon
2. the PSF used in my decon
3. the deringing mask used in my decon
4. the Lum mask used in my decon
5. the post decon image (50 its)
6. the post decon image with a masked conv of the stars to get rid of the bright pixels

[RickS]

Peter,

I grabbed a copy of the files and will have a closer look at them this evening unless someone beats me to it...

Cheers,
Rick.

[peter_4059]

Thanks Rick.

[RickS]

Hi Peter,

I can't start by reproducing your results because the Pre_Deconv image has a different crop to the masks and the PostDecon image, but... I think the problem you are having is just that you're trying to do a heavy decon on the nebula and this is too heavy on the stars (a perfectly deconvoluted star will become a point source.)

If you want to go that heavy then you'll need to do something like Lee suggested and remove the brighter star cores from the luminance mask. You could subtract a suitably constructed star mask or use PixelMath, e.g. "iif($T>0.6,0,$T)" followed by a slight Convolution blur to take out the star cores.

Cheers,
Rick.

[troypiggo]

There. I fixed it.

[RickS]

Nah, you overcooked it Troy :lol:

[peter_4059]

Troy, I was going to say it's a bit green but I guess that would make it undercooked.

[RickS]

Hi Peter,

I just had a similar decon problem. I solved it by using a conventional SNR mask and running a small decon (10 iterations) then I clobbered the stars in the SNR mask [PixelMath: iif(star_mask>0.2,0,$T) followed by a gentle Convolution blur] then hit it with a heavy decon (100 iterations.) That has done the trick and given a gentle star reduction without making them too crispy and still done a strong decon on the nebulosity.

Cheers,
Rick.

[troypiggo]

Can you elaborate on a "conventional SNR mask" please?

[RickS]

A black clipped luminance mask just exposing the bright areas.

[peter_4059]

Thanks for the tips Rick. I collected some more data last night so have had another go at this from scratch. Still getting the bright cores in the stars from decon but have applied some convolution with a star mask and this seems to tame it a bit. There is a lot of Ha in this region!

Thanks for looking.

[troypiggo]

Cheers mate. Figured it was some kind of lum mask.