Outline of object recognition
<p>Nov 01, 2012 09:27 PM
 
look at Feature-based methods
 
find feasible matches between object features and image features...
 
single position of object must acct for all matches...
 
use  linear edges, corners, surface patches
 
2)  hypothesize and test
 
USE backprojection to  generate a rendering of an object...
compare rendering to the image... if similar, accept hypothesis..
 
obtain hypotheses by pose consistency.... alignment
 
a small number of corespondes yields the object position...  others must be consistent...
 
then we can recover the missing (occluded parts of the object)
 
SIFT  scale-invariant feature transform...
 
keypoints of objects are first extracted from a set of reference imags and stored in a DB.
David Lowe 2004...
 
 
 
 
 
 
 
 
 
 </p>