FWHM or Eccentricity when using Subframe Selector

[alanrock]

When using the subframe selector to choose the best sub to use in BPP, which statistic is recommended for choosing the best sub? It seems like it could be FWHM or Eccentricity. Also, once you choose the best one, is there a recommended delta from the best statistics outside of which you'd reject the a frame from the BPP script? Is this based on the MeanDeviation?
Thanks!
Alan

[RickS]

Hi Alan,

To pick the sub to use as a reference for registration I sort the subs in FWHM order and look for a few subs with low FWHM and also low (relative to the other low FWHM subs) Eccentricity.  Then I open the shortlisted subs and inspect visually for any issues that the stats didn't pick up.

I don't think you can make simple rules for deciding what to reject.  It depends a lot on the quality of the data.  I use the SubframeSelector process and check the graphs for FWHM, Eccentricity and SNRWeight and pick rejection parameters so that the worst outliers get tossed out but there's still plenty of data left to work with.  If I have a lot of subs then I'll be more picky than if I only have a few.  I often calculate a weight for each sub based on a combination of FWHM, Eccentricity and SNRWeight and use that for integration weighting rather than the default.

Cheers,
Rick.

[dld]

Hello Alan,

You may use 1/(FWHM*Eccentricity) as a weighting expression for picking the best reference frame in terms of "sharpness". It has units 1/pixel (if you are working with pixels instead of arcsecs) and weighs higher the images with round and small star profiles.

If you want to use the weights for integration purposes, you may need to include SNRWeight somehow.

Writing weighting expressions depends on the situation (how much data, light pollution) and what you're aiming for. For example you may weigh for sharpness if you're imaging a globular cluster and be strict with your rejection threshold, or for SNR if you're imaging some faint dusty thing and be less strict with your rejection threshold (because you don't want to toss away integration time).

Personally I give a moment to think of the units of my expression, and how it behaves when a factor takes high/low values. I also like to weigh unit-less things for example,

weighting_expression = weight_1 * unitless_sharpness + weight_2 * unitless_SNR,

with weighting_expression > 0 because later in your processing ImageIntegration will fail with negative weights. I hope this makes sense!

Happy weighing