The  Weizmann  Institute  of  Science
                  Faculty of Mathematics and Computer Science


                          Vision and Robotics Seminar

                              Michel Vidal-Naquet



                                 will speak on


                   Visual Features for Object Classification

Abstract:
In this talk, I will present and discuss advantages of a class of visual features, called image fragments,
for the task of object classification. Fragments are class specific image patches of intermediate size,
selected to maximize an information criteria about the class. I will first show how an optimal set of
fragments can be found, and compare the fragments in terms of information content about the class to commonly
used features, such as back propagation receptive fields and Non-Negative Matrix Factorization features.
I will then discuss the use of fragments in classification tasks and show that the choice of informative 
features can simplify the complexity of classification and learning.





                      The lecture will take place in the 
                     Lecture Hall, Room 1, Ziskind Building
                         on Thursday, January 30, 2003
                                  11:00 - noon