Subject: 
       Georeferencing
   Date: 
       Tue, 18 Feb 2003 13:17:55 -0700
  From: 
       Jim Heffernan <jim.heffernan@asu.edu>
    To: 
       Chris Eisinger <chris.eisinger@asu.edu>,
       John Roach <john.roach@asu.edu>,
       Tyler Rychener <Tyler.Rychener@asu.edu>
    CC: 
       Ramon Arrowsmith <ramon.arrowsmith@asu.edu>




Hi Guys-
 
Unfortunately I didn't make a great deal of quantative progress yesterday.  However, I think I did learn a few things that
hopefully will be of use.
 
1) I didn't crop the pictures enough.  take an extra 10-20% off the edges before you bring the photos into ArcGIS.  This
will reduce error and the photos still overlap.
 
2)  It is difficult to get RMS under 8 m while still having even coverage.  However, around 8 RMS the 1962 photos are at
least as well-matched to the DOQQ's as are the topo maps. Perhaps more experienced folks could do better.   
 
3) For me, RMS Error is more a function of point selection than # of points, at least for any reasonable number.  Beyond
12 points
 
Here is a rough outline of the procedure I think would work and would balance speed and accuracy
 
1) Establish control points at all four corners
 
2) Choose a point in the center of the map
 
3) Choose 4 additional points near the center of each side.
 
4) Steps 1-3 should give you a reasonably even spread of points across the photograph.  Now check the links file.  Do any
of the points have particularly high residuals?  If so, this may be a bad point.  delete the point and use another control
point in the same region.  With 9 points, you should be able to get a good RMS if the points are well-chosen.
 
Feedback on this procedure is welcome.
 
Jim