RE: Focus indication

[<aah@nofs.navy.mil>]

Rob wrote:
>Anyone have thoughts on how to determine when the "background tail is
>wagging the star-dog" for Michael's method?  Since I'm not going to present
>information on an individual star basis, but rather a mean of a subset of
>behaved data, this might not be an issue.
  As I suggested to Tom, an automatic determination of fwhm for a frame
is not a trivial matter.  Mike mentioned the variation of the psf across
the field.  Many of the frames have thousands of stars; you have to
somehow select which subset you want to use for your 'behaved' data.
Because of the many stars, the background is *not* well behaved, and
an accurate determination of the sky level is not easy.  You have to
avoid saturated stars and also stars with low signal/noise.  You have
to avoid dots on airplane and satellite trails, CCD cosmetics, etc.
You have to avoid stars that are blended.
  For our non-automated system at the 1.0m, we hand-pick a single star
in a frame for which to acquire statistics at the end of an exposure.
That fwhm is reported in the computer-generated log file.  For my
pipeline, I determine fwhm for all stars, then start paring based on
the points mentioned above, and finally determine the modal value for
the remaining stars and report that value in the extracted starlist header.
  If Tom really wants an automated fwhm report that he can trust, then
you have to do a fair amount of coding.  Pick only the central part of
the image to avoid the coma problems.  Then you have to somehow find
all stars within that window.  Reject all stars with peak DN
above the saturation level.  Reject all stars with peak DN less than
some threshold.  Check to see if any stars remain!  Then find fwhmx,y
for the remaining stars.  Find median fwhmx,y.  Reject stars that are
obvious outliers (probably blends).  Find median fwhmx,y of the remainder.
Once you have done all of these steps, you are almost at the full pipeline level.
  If this goes into a log somewhere, then the reader will have to hand-check
files with deviant fwhm to see if it is truly deviant or whether the
automated code had problems.  While the determination of fwhm is
useful at the beginning of a night to ensure everything is working
correctly, I fail to see the value for such a feature for the rest of
the automated night.  The data will be fed into a pipeline shortly
thereafter and accurate values obtained; a night with problems will be
indentified at that step.  Perhaps I am not understanding the problem
and someone can give me some reasons.
Arne