Appearance-besed object recognition

                           Jan Burianek

Appearance-based object recognition is a novel approach to object
recognition.  The distinguishing features of this class of recognition
methods are:

  1. Use of a representation derived from 2D views of objects
     instead of an explicit 3D model of an object.  

  2. Models of objects are acquired (typically automatically) from 
     training images using statistical learning methods.

  Basic definitions, motivation and properties of appearance-based
recognition will be presented. Previous work will be reviewed. Four
examples of A-B methods will be discussed in detail: direct search
(Nayar) and wavelet based methods (Rao, Ben Arie, Wu).