Using non-perfect subs in stack?

[Datalord]

44 minutes ago, carastro said:

So a total of 55 hours in all? 

Yes. And completely removing the stars from the bad is a tactic I'll try along with Vlaivs. 

Edited July 17, 2019 by Datalord

[ollypenrice]

11 minutes ago, Datalord said:

Yes. And completely removing the stars from the bad is a tactic I'll try along with Vlaivs. 

Hmmmm... everything else is as trailed as the stars... This is a nebula containing some very fine structures. Still, it takes next to no time to run a stack so a bit of experimenting might well pay off.

Olly

[Rodd]

I have found the following

1) PI does a remarkable job of removing satellite and plane trails--might have to tweak settings for bad cases, but I have never had a plane or satellite trail that showed up in final image.

2) As long as the subs are not "terrible" I find that including less than perfect subs in a stack is not really noticeable in final result--as long as the subs are not too bad.  I usually exclude subs based on FWHM, Median and SNR.  Sometimes SNR will decreases from say 1.1 to .6 in sub frame selector and it looks like a huge drop--a big spike downward--but the subs look pretty much the same if you open them.  These are OK.  But sometimes a cloud is passing by and the sub is greyed out (its Median value will be very high)--this HAS to be removed or it will ruin the image.  Sometimes there will be a huge spike upward in SNR--this is bad too--it means a cloud is passing by and for some reason PI equates it with signal.  At the very least, the subs that have a wildly different SNR or Median should be inspected.

I have integrated 100 subs, then went back and took out the 20-30 worst (FWHM twice the others, and poor SNR).  The final stack was virtually identical to the one with all subs.  The FWHM was .2 larger (Not significant for me).  SNR is always much higher with all subs--but side by side--its tough to see the difference.  However, differences may reveal themselves during processing--higher SNR images can be sharpened more, and other things.

so--A spike in Median--always remove, means cloud

A spike in FWHM--as ling as it is not huge, or the result of a star trail or glitch in guiding  (stars with multiple centers or box apperances), and as long as its not a lot of subs-keep

Spike (upward or downward) in SNR--inspect image.  most times, as long as the Median does not spike as well, keep

Rodd

[Datalord]

Ok, started with a full integration of all Ha frames I had. 64 frames of 20 min. Used PI Subframeselector to give weights to the images with a weight formula heavily, heavily focusing on eccentricity and FWHM: (20*(1-(FWHM-FWHMMin)/(FWHMMax-FWHMMin)) + 80*(1-(Eccentricity-EccentricityMin)/(EccentricityMax-EccentricityMin)) + 1*(SNRWeight-SNRWeightMin)/(SNRWeightMax-SNRWeightMin))+50

I threw away 0 frames in this "junk" run. Then I normalized and stacked and it gave me this:

That's so good I was very pleasantly surprised. I looked closer and not only were the stars tight, they were even kinda blocky, so I tried to drizzle2. You can see the normal stack left and the drizzled on the right.

Not much difference at this scale, but closing in on the pixels the drizzled stars becomes more round in their white profile.

I haven't stacked the "good" stack alone, but that is the next task. It will have to be pretty amazing to convince me not to keep this one, though.

[Datalord]

Got around to stacking the "Good" images. 37 of them, clipping off the worst eccentricity and FWHM. Then I normalized, stacked and nuked it with the exact same ScreenTransferFunction as I did with the full junk stack.

Left is "junk", right is "good".

I can't see much difference. If anything, there is a bit better control over the noise in the "junk" stack.

[ollypenrice]

10 hours ago, Datalord said:

Got around to stacking the "Good" images. 37 of them, clipping off the worst eccentricity and FWHM. Then I normalized, stacked and nuked it with the exact same ScreenTransferFunction as I did with the full junk stack.

Left is "junk", right is "good".

I can't see much difference. If anything, there is a bit better control over the noise in the "junk" stack.

I agree. Precious little in it but the darkest parts are a tad smoother in the full set. Good experiment and well worth the effort.

Olly

[kirkster501]

Pixinsight subframe selector can weight the subs such that ones less than a given FWHM combine less to the final integrated image.

I maintain that if you have subs where you can visibly see that there is some star trailing when you examine them then you have to ditch those subs, painful though that can be.  It is a result of this that I go no higher that 10-15 minutes in my skies per sub.  20/30 minute subs (which my rig is capable of) is too much time invested on a precious clear night on one sub that can go wrong. 

I *do* keep subs where they were slightly misted out due to a bit of high cloud passing through, though.

[Datalord]

31 minutes ago, kirkster501 said:

I maintain that if you have subs where you can visibly see that there is some star trailing when you examine them then you have to ditch those subs, painful though that can be.

I think my experiment above shows you should actually keep them and use an aggressive weight formula in the SubframeSelector. I'm quite pleased with the results I got. 

[eshy76]

Just like in PI, APP (Astro Pixel Processor), allows you to weight subs by quality, including FWHM, when stacking. There is even a control to choose the percentage of subs to stack, so if you think that 10% of your subs are potentially bad, you would select 90% and would get the best 90% subs weighted by quality.

This is a faster process than PI I find, as it's two settings then click, but you can always inspect the sub stats after weighting and manually remove ones you don't want in the same way if you wish.

In your position that's the approach I would take and see if I was happy with the result.

[Datalord]

More followup to some of @vlaiv comments about PSF. I started digging into the processing of the "junk" stack I drizzled(2) and before deconvolving I have to create this PSF from the image.

I'm honestly shocked at how round it is. Remember, this includes all those hours of "bad" images; I believe I have 12 hours of "good" Ha and 11 hours of "bad" that went into this stack.