00100	INTRODUCTION
00200	
00300	       Verification vision  can be roughly  described as the  process of
00400	matching  a predicted image or  model with an actual  image when the two
00500	are nearly the same.  In particular, the identities of the elements (eg. 
00600	bolt,  corner, or  correlation patch)  are not  in question;  only their
00700	relative  positions (and possibly the  confidences associated with these
00800	positions). 
00900	
01000	       Verification vision has been  used in various ways in  the past. 
01100	Possibly  the most common  is "hypothesis  and test." For  example, some
01200	higher level procedure proposes  a certain line  (limited to a range  of
01300	angles)  at  a certain  place  in  a  picture (within  tolerances);  the
01400	comparison  step is  supposed to locate  the line,  if it  is there, and
01500	return the  matching angle  and  position (see  [FALK], [SHIRAI],  ...).
01600	Another place  verification vision has been used  is within narrow-angle
01700	stereo programs.  In this case the "prediction" is another image of  the
01800	same objects,  but taken  from a  slightly different relative  position.
01900	The goal is  to locate matching "features" (such as correlation patches)
02000	in order to provide the stereo  package with two positions for the  same
02100	part of the scene (see [Thomas]).   Notice, as mentioned above, that the
02200	identities of the models (ie. the line and correlation patch) are not in
02300	question; only their positions in the actual image. 
02400