Re: FITS tables (really sharpness and roundness)

[Stupendous Man <richmond@a188-l009.rit.edu>]

  Jure asked:

> How is the roundness defined? ... 
> how exactly is the sharpness defined? 

  I use the definitions put forth by Peter Stetson in his first
article describing DAOPhot, which is in Publications of the Astronomical
Society of the Pacific, vol 99, p. 191, 1987.  You can find a copy 
on the ADS server, specifically (entire this as a single line in your
browser)

     http://adsbit.harvard.edu/cgi-bin/nph-iarticle_query?
                bibcode=1987PASP...99..191S

  He says:
                       ratio of central pixel to nearby pixels
         sharpness =  ----------------------------------------
                        height of a fitted stellar profile

Defined in this way, true stars will have a value of around something
less than 1.0 and bigger than 0.1, cosmic rays values bigger than 1.0,
and noise peaks values close to 0.0.

  One often finds this corrupted to something similar, such as
"ratio of the central pixel value to nearby pixel values".  For some
purposes, this is okay; it still serves as a diagnostic to separate
stars from not-stars.

  Stetson defines roundness as 

                           (hy  -  hx)
          roundness = 2 * -------------
                           (hy  +  hx) 

where "hx" is the height of a 1-D fit to an object's profile in the 
x-direction, and "hy" is the height of a 1-D fit to an object's profile
in the y-direction.  Bleed trails cause a big fit in one direction
(parallel to trail) and a small fit in the other direction (perpendicular
to trail), whereas true stars have roughly equal values in both directions.
Thus, the roundness is something like 0 for true stars, and something
like +/- 2 for bleed trails or bad columns.

  Again, one can substitute some similar measure of "image size in
x-direction versus image size in y-direction" to separate stars
from noise, and call it "roundness".

  The basic idea here is that, regardless of the exact definitions,
if you are handed a list of detections from an image, and you see
that MOST of the objects have, say, sharpness=0.65, but a few objects
have, say, sharpness > 1, then it's a pretty safe bet that those few
objects aren't true stars, or are significantly contaminated by some
sort of noise.

  If you plan to examine stars in a SINGLE star list, it doesn't really
matter how these terms are defined -- you can pick out the good from the
bad.

  If you plan to combine measurements in MANY star lists, created by
many different observers, then it _does_ matter: if 3 different people
define "sharpness" slightly differently, and throw all their data 
into a database, then it won't be correct to make a simple selection
such as "sharpness > 0.4".  Instead, you would have to make separate
selections for each person's data.  Ugh.

  Quick note on FWHM: if it were up to me, I'd compute the FWHM along
the x- and y-axes, rather than along the major and minor axes.  It's just
easier, and (most of the time) will still let you know when things are 
trailed badly.

> Also, I will repeat my question from one of the previous emails.  Is it a
> realistic scenario to shuffle half-processed star lists with millions 
> of stars all around the planet until they are well analyzed or is 
> it better to do complete processing at the site which produced the data?  

  My preference is for each site to complete the processing; less work
for me (and everyone else) to interpret the results.  I do understand that
this requires extra software to do the astrometric and photometric 
calibration.

                                            Michael Richmond