PI subframe selector formula

[ChrisV (Chris)]

I've got a bit of time to kill today so had a look at the new PI features. I'm liking the new subframe selector 'process' in pixinsight. But its made me think about how I calculate weighting for approved lights. In the past I've used a weighting based on the David Ault spreadsheet - a combination of FWHM, Eccentricity & SNRWeight and varying their relative contributions some times. All good.

But now I wondering how much I should scale these overall. At present I'm using a weighting range from 40-100%. So for example - normalised FWHM 35% + Eccentricity 7% + SNRWeight 18% + 40%.

I've noticed some use a range from 0 (no contribution) - 100 %. Given that I'm using a CMOS camera and churn out a few hundred subs, maybe I would be better off being more aggressive -using a 0-100% weighting range ? Any thoughts?

[RickS (Rick)]

My starting point is:

80*(1/(FWHM*FWHM)-1/(FWHMMax*FWHMMax))/(1/(FWHMMin*FWHMMin)-1/(FWHMMax*FWHMMax))
+15*((SNRWeight-SNRWeightMin)/(SNRWeightMax-SNRWeightMin))
+5*(1-(Eccentricity-EccentricityMin)/(EccentricityMax-EccentricityMin))

Which weights FWHM heavily.

Cheers,
Rick.

[multiweb (Marc)]

Originally Posted by RickS

My starting point is:
Which weights FWHM heavily.
Cheers,
Rick.

Sure you didn't come up with that one.
I found it in the Michelin Guide for their satisfaction rating:

80*(1/(PORT*PORT)-1/(PORTMax*PORTMax))/(1/(PORTMin*PORTMin)-1/(PORTMax*PORTMax))
+15*((CHEESEWeight-CHEESEWeightMin)/(CHEESEWeightMax-CHEESEWeightMin))
+5*(1-(Cracker-CrackerMin)/(CrackerMax-CrackerMin))

[RickS (Rick)]

Originally Posted by multiweb

Sure you didn't come up with that one.
I found it in the Michelin Guide for their satisfaction rating:

I like the way you think, Marc

[ChrisV (Chris)]

Wow that's all or nothing - and with the ^2 that's a full on cut-off (I think). No room for in between, even with crackers.

Why the emphaaasis on FWHM versus SNR? I noticed that the new process also has 'noise' but I suppose SNR takes care of that anyway.

[Stonius (Markus)]

Have you tried asking on the PI forum? Might get deeper specialist knowledge there.

Also, I'd quite like to know the answer myself.

Markus

[RickS (Rick)]

Originally Posted by ChrisV

Wow that's all or nothing - and with the ^2 that's a full on cut-off (I think). No room for in between, even with crackers.

Why the emphaaasis on FWHM versus SNR? I noticed that the new process also has 'noise' but I suppose SNR takes care of that anyway.

I do a noise weighted integration as well and compare the results. The noise weighted result always has better SNR but often not enough that you'd notice after a little processing. Some times the FWHM improvement isn't that great and I just go with the noise weighted integration.

I have also had some success combining the bright parts of the low FWHM integration with the dim parts of the noise weighted integration for a best of both worlds result.

[ChrisV (Chris)]

Thanks Rick. Very informative- "deep specialist knowledge" as usual. I like the combo of bright and dim parts! There's just so many potential paths - but only one life to try it all ...

I must go back and compare it to plain noise evaluation.

And Markus, I've put it in the PI forum.

[RickS (Rick)]

Chris,

Something else to play with is including or excluding some of the lower quality subs. I always dump anything cloud affected but I tweak the acceptance criteria a bit depending on whether data is scarce and SNR is lacking or I have plenty and I'm aiming for the sharpest result. The SubframeSelector process allows you to base criteria on the standard deviation of different parameters which is handy, e.g. "FWHMSigma<2.5".

Cheers,
Rick.