RE: troubles with matching

["Creager, Robert S" <CreagRS@LOUISVILLE.STORTEK.COM>]

Hey Folks,

I have written a perl script which uses Michael's match program to search a
given ra/dec range for matches from a reduced star list.  It's pretty
simple.  Given an offset in both ra and dec (offset from the picture
ra/dec), it will apply the match program multiple times to zero in on the
'best' solution.

I've had great success with it so far.  I've also written a program which,
if I cannot find a match for one band, but the other does match, I match the
two frames together and then apply the two transformations (x/y then ra/dec)
to the non-matching frame.

On the 'back end', it requires a seperate script to spit out the catalog
stars with a given range in a specific format.  I use a db with script
interaction for this, but a flat file could be substituded.

I'd be happy to supply the perl script I wrote, and some enterprising coder
could fit it into Michael's pipeline (in C I would presume).  The script
(and a support package) could be used as is, as all aspects are configurable
(ra/dec columns, algorithm parameters).  I don't have the script here at
work, but can grab it tonight.

Cheers,
Rob

> -----Original Message-----
> From: Stupendous Man [mailto:richmond@a188-l009.rit.edu]
> Sent: Monday, April 01, 2002 11:42 AM
> To: tass@listserv.wwa.com
> Cc: mwrsps@rit.edu; Richard.Knowles@corporate.ge.com;
> tdroege2@earthlink.net
> Subject: troubles with matching
> 
> 
> 
>   Tom and Rich write:
> 
> > Rich Knowles and I are struggling with a pair of images.  
> The V image finds 
> > .ast files perfectly.  The I images will find only one or 
> two, and juggling 
> > values in make_list.out might make one image convert and 
> another disappear 
> > from the .ast list.  We only get one or two successful out 
> of 56 with all 
> > 56 V images successful.  I had this problem on DS20 7 and 8 
> and was able to 
> > fix it by modifying make_list.out.
> 
>      [brief digression: DS 20 disks 7,8 have RA = 71.2, Dec = 7.4]
> 
> > OK, the thing special about this list is that it is taken 
> at the equator 
> > and is thus dense. 
> > ...
> > What values do we fuss with to try to get .ast files?  Can 
> you give us the 
> > values to change and some logic behind them?  Possibly we 
> can talk Rich 
> > into working on any code that is necessary to help make the 
> process run 
> > smoother.  How close do we have to be to get a match? 
> 
>   This is a complicated problem.  There is no easy answer.  
> 
>   The problem is (probably) that there are a LOT of candidate stars
> detected in the image.  The reference catalog file is based on the
> Tycho-2 catalog, which contains stars with magnitudes mostly 
> V=8 to V=10.
> The density of stars isn't very large: there are about 353,000 stars,
> so, if they were spread evenly across the whole sky, there would be
> only about 8 per square degree.  Since the Mark IV field 
> covers about 16
> square degrees, one expects around 100-200 catalog stars per 
> Mark IV field.
> 
>   So, there are thousands of stars detected in the image, and 
> only one or
> two hundred catalog stars.  How can we match them up?  The pipeline
> is designed to
> 
>        - pick out the brightest 300 (by default) stars from the image
>        - pick out the brightest 200 (by default) stars from 
> the catalog
>        - run the "match" program on these two subsets
>     
>   If there are SO MANY stars in an image that the brightest 300 stars
> in the I-band, say, are largely brighter than the stars in the
> reference catalog, then there might only be 20 or 30 true matches
> between the two subsets.  The "match" program is likely to fail
> if the number of false matches is much larger than the number of 
> true ones.
> 
>   There are several parameters which one can set in the 
> "match" program
> which may help it work in these fields.  You can read about them
> in the "match" manual
> 
>           http://spiff.rit.edu/match/match-0.5/match.html
> 
> The ones most likely to help are
> 
>        - trirad  (try making it smaller than the default)
>        - scale
> 
> It would take me too much time to explain exactly how these are
> calculated and used.  You might try reading the manuals, and then
> the code in "astrom.tcl".  But you could try fiddling with these
> slightly to see if some particular combination helps the code
> to discard false matches and keep only the good ones.
> 
>   Brave people may edit the "atpmatch.c" file in the "match" directory
> and change the #undef DEBUG to #define DEBUG, then recompile.
> The program will print a LOT of diagnostic messages to stdout as
> it runs.  It will confuse the pipeline (which doesn't expect to see
> all the extra messages), but allow a human to figure out why the
> stars aren't matching properly.
> 
>   It may help if I modify the pipeline so that instead of 
> using the brightest N stars from the reference catalog, it 
> instead uses 
> a set of N stars which lie between two particular magnitude
> limits.  Unfortunately, we can't constrain the detected stars to
> lie within a particular magnitude limit, because we don't (yet)
> know the zero-point offset between the instrumental magnitudes
> and calibrated magnitudes when we start the matching process :-(
> 
>                                             Michael
> 
>