Is FWHM the best measure to use?

I don't recall ever seeing details of the algorithm that APP uses to calculate quality but it would be most surprising if FWHM was not a contributing factor in the calculation along with #stars, dispersion, registration RMS, etc measurements that APP calculates along the way.  The Quality score therefore should be a more robust measure of overall image quality than any of the others individually.  

I can valuely recall processing images which looked good on the FWHM measure but visually did not look so good. After checking I eventually put this down to the images being captured during a hazy interval such that only light from the brighter star cores was detected. These images as a consequence I think scored less well on dispersion and #stars, resulting in I think a correctly lower quality score.

My advice therefore would be to take a quick look at any image files that score exceptionally well on any single factor (eg FWHM) but overall obtained a poor quality score.

Also do not deselect image files until after normalisation after which APP has the  fullest information set available for its quality calculation.

The only times I can recall APP calculating what I thought were unwarrantedly high quality scores were a few occasions when I think my telescope experienced a jolt or tracking glitch during exposure such that APP was presented with an image file with double star images and thus achieved incorrectly high star counts.

My thoughts for what they are worth.

Mike

PS Maybe Mabula could start a topic where users could report what they think were seriously miscalculated quality scores for analysis and maybe eventually an  improved quality scoring algorithm.

I always run into the question of when and how many poorer quality frames to exclude from integration.  I may be wrong, but I seem to recall that APP will progressively use less of the poor quality data from lesser quality frames. But the final image will somehow benefit from those frames' "good" data for an overall better final image.  Is that true?

'I may be wrong, but I seem to recall that APP will progressively use less of the poor quality data from lesser quality frames'

I am sure there must be a better explanation somewhere in the forum but this is my understanding also, see below screen shot.   As I now tend to use 'automatic' integration mode these tips are not shown so much but I don't think the underlying use of weights (quality or otherwise) during integrations has changed.

So when using the Automatic stacking setting, is the recommended best practice not to take time to deselect lower rated Quality images before integrating?  Except maybe the really bad ones?

Hi @dalemschultz, @mestutters, @jhart

Depending on the dataset, you want to create analytical charts for all the analysed metrics after step 5) normalize, so make charts for the star size, FWHM, noise, background, etc, and check if some of these charts have clear outliers on the negative side. These are the lowest in the charts. If they are really outliers, it will best to remove them.

In our experience on many datasets, you want to be careful with removing frames, if you are too aggresive and remove too much, your result will be less good in terms of Singal To Noise Ratio. So yes, if you keep some of the bad frames, the result can still be better. Be carefull and after integration, check the result visually of course but also check the analytical results in the FITS header of the integrated file, you can find the noise, SNR, noise reduction etc... there 😉

Mabula