Application of Fisher information to detection problems in computer vision

Speaker:          Dr. Steven Maybank 
                  The University of Reading, UK  

Title:            Application of Fisher information to detection problems 
                  in computer vision 

Date:             Monday, 22 September 2003 

Time:             4:00pm - 5:00pm 

Venure:           Lecture Theatre F (Leung Yat Seng Lecture Theatre) 
                  (near lift nos. 25/26)   


ABSTRACT: 

The framework for many detection problems includes a parameterised
family of probability density functions theta -> p(x|theta) where
theta is a parameter and x is a measurement. The task is to find a
value of theta compatible with a given set S of measurements. Examples
of such detection problems include the detection of lines, circles and
ellipses in images, the detection of the epipolar transform between
pencils of epipolar lines and the detection of the collineation
between two images of a plane.

An asymptotic approximation is obtained for the Fisher information of the
family of densities theta|->p(x|theta). Under the Fisher information the
parameter space becomes a Riemannian manifold. These results are the basis
of new algorithms for detecting lines and epipolar transforms.


BIOGRAPHY: 

Dr SJ Maybank obtained a BA in Mathematics from King's College,
Cambridge, UK in 1976 and a PhD in Computer Science from Birkbeck
College, University of London in 1988. He worked as a research
scientist at GEC (UK) from 1980 to 1995 and joined the Department of
Computer Science at the University of Reading as a lecturer in
1995. He became a Reader in 1998. Steve Maybank has carried out
research in nearest neighbour pattern classification, the geometry of
multiple images, applications of invariants to computer vision, CCTV
surveillance and applications of Fisher information to computer
vision. He has published over 80 scientific papers and one book.