Re: Minimum spanning circles and Voronoi diagrams

[Chris Albertson <chrisalbertson90278@yahoo.com>]

What is the source of astrometric error bars?  Isn't it
a function of the size of both the PSF and the pixels?

Also, I think if you take Tom's suggestion and plot the
data. (Draw a one sigma circle at each detected star's
location) I think you will see that a simple threshold
would work.  We did this for Mk III data and found that
it worked well enough.  The stars are something like
8 or 10 sigma apart because we can centriod them
to within a sub-pixel but can't detect them unless
they have a few pixels between.

"Creager, Robert S" wrote:
> 

> Let me try to restate this, as I understand what you're saying, but am not
> sure how to get the data to apply.  You choose some radius to match stars
> to.  If that radius is too small, you may end up with two stars (the same)
> very close.  You plot the sigma's addictively (where they intersect), and
> based on some saddle height criteria, you know only one star exists.
> Correct?

I think that's it.  My thinking was that if the minima of the
sum of the two functions is >2 then the one sigma rings must
intersect.  Turns out it's not true because 0.5 + 1.5 = 2.0
But still, I think you only need to evaluate the distribution
function along a line conecting the two stars.

> > Still, I'd just compute the angle between the stars and
> > apply a threshold.
> 
> But, like I said, this is computing the angle between all the observed stars
> which went into making one star, right?

OK, I finaly caught on.  You are solving this for a group of N
stars.  I did it diferently.  I had a set of observations that I
called "one star"  I defined the location as the mean of all
observations in the set.  My problem was to decide it the
next observation was to be added to the set or if it was to
be the start of a new set.  If I added it I'd update the
location of the set.  Over time the mean location stayed
put.

The reason I did this was because I did not want to have to
read up all that historic data every time one new point
was added.  I could update the running count, means and sigmas
without accessing all the old data points.  Much faster.

Your method could be better but what? 50X slower?  I think in
a production evironment, use my faster method then come back
later and use the more powerfull technique where there is
some indication we have a double.
- 

--
   Chris Albertson             
   Redondo Beach, California
   home:   310-376-1029   chrisalbertson90278@yahoo.com
   cell:   310-990-7550
   office: 310-336-5189   Christopher.J.Albertson@aero.org