Re: Minimum spanning circles and Voronoi diagrams

[Chris Albertson <chrisalbertson90278@yahoo.com>]

It depends on the algorithm used by the software.  Here is
another scenario:

The image in convolved by the system PSF the peaks of the
convolution are found and put in an internal table.
Thus, only "star-like" objects will be found.  Isolated
bright pixels don't make tall peaks in the convolution.

For each peak:

Fit a model PSF by scaling a model PSF by a multiplier. Adjust
the multipier for least squars best fit.  Next substract the
scaled model from the image.  Look at the remainder.  The final
value of the multiplier is the "brightness" and the 
remainder is the "error"

Yes I think the cosmic ray will pull the least square fit up
a little but the remainder will be larger and show up in the error
bar.

You double that is resolved on some images and not others is
a real problem.  We saw this with the Mk III.  I think the
best we can do is to look periodically at the database and
try and find them.  You need many observations before the
statistics are there to make a problem show up.  I think
Michael wrote some software that scans a database looking for
groups of observatins to be split or merged.  Maybe now we
could trigger this kind of check automatically after the 
(say) 20th data point is added or if the sigma of ra or dec
goes over some threshold


--- Tom Droege <tdroege2@earthlink.net> wrote:
> Chris,
> 
> OK, consider the following situation.  A cosmic ray hits slightly off
> 
> center for a star.  This increases the number of counts in the star,
> and 
> shifts its position toward the cosmic ray.  Since the total
> photometry has 
> more counts in it, the error computed is *smaller* than if the cosmic
> ray 
> had not hit.  Further, the photometry is off, the star is measured as
> too 
> bright.  You can't just use the error that comes out of the 
> photometry.  You have to know the other information.
> 
> In this case, the only thing you know is that this position
> measurement of 
> the star is different in position from most of the other measurements
> of 
> the star.
> 
> Another example.  A little high thin cloud will spread out the star 
> psf.  Now two close stars that might have been separated previously
> will 
> appear to be one star.  Don't argue that this can't happen.  There is
> some 
> set of positions for the two stars where a little cloud will cause
> stars 
> that were previously separated to blend into one.  Again the blend
> will 
> cause the position to change over the clear sky case.
> 
> I think the error is not enough.  We will have to carry some markers
> along 
> with each measurement.
> 
> Tom Droege
> 
> At 08:38 PM 2/6/02 -0800, you wrote:
> >Tom Droege wrote:
> > >
> > > I cannot imagine that one can do as well "deblending" as one does
> with a
> > > clean image.  So I would want to mark such measurements as "not
> first 
> > quality".
> >
> >I'm not 100% sure but I think the mark of "not first quality" is
> >a larger error bar.  The neat thing about using a database is
> >that you can specify a selection criteria when you take the
> >data back out.  We just need to keep processing history and error
> >bars for every data point.  We did this for thr Mark III data
> >mostly.  I see Mk IV as just "much more of the same".
> 


=====
Chris Albertson 
  Home:   310-376-1029  chrisalbertson90278@yahoo.com
  Cell:   310-990-7550
  Office: 310-336-5189  Christopher.J.Albertson@aero.org

__________________________________________________
Do You Yahoo!?
Send FREE Valentine eCards with Yahoo! Greetings!
http://greetings.yahoo.com