Comparison of ImageIntegration Results via SubFrameSelector Script

[jerryyyyy]

Had the idea of comparing various methods of integration using the SubFrameSelector script.  I save the variations then evoke the script on them.  This gives an Number of useful parameters to compare.  But I know that math of this is tricky.  Does this make any sense....  It is very convenient.,

[RickS]

Jerry,

I have been doing this for a while and I think it is a useful technique when used carefully. In particular I think you need to be cautious about the SNR estimates.

Cheers,
Rick.

[jerryyyyy]

Yeah, I am not sure how it is doing the SNR calculation, but since that forms a lot of the basis for subframe selection, either there is something to it or subframe selection as a script is not valid. 

[pfile]

well it could be that SFS is expecting subframes rather than integrations - the statistics of a subframe are going to be different than an integration, which could render the analysis less than valid. i guess mike would have to comment on the script's accuracy on integrated images.

[jerryyyyy]

FWHM and Star Support should be unique for each image, I would think.  I think you would get into trouble with the Sigma values that I assume get a SD off a collection of images. 

I found the write up of the calculations for the noise etc:

https://pixinsight.com/doc/scripts/SubframeSelector/SubframeSelector.html#__description__

Seems most of the parameters are straightforward and come off PSP (DynamicPSP used in deconvolution).

Finishing my travel today and will be integrating some new images.  I can compare Linear vs Winsor then compare manipulation of some of the cut offs.  Also can compare rgularL versus Super L. 

[jerryyyyy]

Thought I would bump myself here to put up a new image.  These are data from the DSW M31 files for L using all data to make Super_L.  There are 4 files: 

Linear Integration 5 and 2.5 versus 5 and 4.0 Hi/Lo and Winsorized 4 and 3 (default) versus 2.9 and 2.9 (A Pfile suggestion). 

According to the figures i see Pfile's settings win.

[RickS]

Thought I would bump myself here to put up a new image.  These are data from the DSW M31 files for L using all data to make Super_L.  There are 4 files: 

Linear Integration 5 and 2.5 versus 5 and 4.0 Hi/Lo and Winsorized 4 and 3 (default) versus 2.9 and 2.9 (A Pfile suggestion). 

According to the figures i see Pfile's settings win.

Well, they win on FWHM and lose on SNR 

Cheers,
Rick.

[pfile]

well you can't really just take those two numbers and apply them to any image - on the one you had showed me there were lots of hot/cold pixels left in the integration and so i thought it was worth a try to try to kick those out with a smaller sigma.

basically you should integrate without any rejection and see what the SNR is (shown in the console at the end of the integration). then start messing with pixel rejection until you're happy with the rejection maps, but keep an eye on the SNR to make sure it has not gone to hell (meaning you have rejected too many "real' pixels)

rob

[RickS]

Also worth trying to get rid of the bulk of your hot and cold pixels with CosmeticCorrection before integration, but as Rob suggested you don't need SubframeSelector to compare the SNR impact of different rejection algorithms and parameters.  ImageIntegration displays the data you need on the process console.

Cheers,
Rick.

[jkmorse]

Rick and Rob are right (of course).  Here is a presentation that explains much of this:  http://www.astrosurf.com/jordigallego/articles.html

Here was my take away from that presentation.  As a first step in Pixel Rejection, before running your selected rejection algorithm, run ImageIntegration with No Rejection selected.  After the run is complete, scroll up a bit in the Process Console readout to see the parameters for the run.  There you will find a number labeled “Reference Noise Rejection”.  That is the ideal you are trying to duplicate with your rejection algorithm.  Anything less and you are leaving excess noise in your image; but at some point you start rejecting valid data.  The key is to find the sweet spot. 

Rerun the tool, but this time with the rejection method selected and set with your starting settings.  After the run, see how close you are to that ideal by again checking the results in the Process Console report.  Tweak the settings (primarily the high setting) until you get as close as you can to that ideal.  At that point you have maximized what ImageIntegration can do for you.  Keep in mind, if you start getting the same or very similar numbers as you set the high or low numbers to values closer to zero (i.e. with greater rejection), pick the highest  settings (those farthest from zero) that are close to the ideal so that you maximize noise elimination while also maximizing retained valid data.

Best,

Jim

[jerryyyyy]

Thanks Guys, really learning a lot.  Jim do you still keep your cheat sheet?  Love to get a recent iteration as I am making my own up with a compendium of wise comments at each stage of processing.  Really need an algorithm  Your comments will help with that pixel rejection magic. 

[jkmorse]

Jerry,

Sure do.  Just drop me a line at jkmorse57@gmail.com and I will send along my latest.

Best,

Jim