COLLOQUIUM
         Computer Science Department, Boston University

Speaker: Vladimir Pavlovic
         Dept. of Computer Science
         Rutgers University
         http://www.cs.rutgers.edu/~vladimir

Title: A Graphical Model Framework for Coupling MRFs and 
       Deformable Models for Image Segmentation

Date:  October 1, 2004
Time:  2pm
Place: MCS 135 (for directions, see www.cs.bu.edu/colloquium)

Abstract:

Image segmentation is one of the most important and 
difficult preliminary  processes for high-level computer 
vision and pattern recognition problems. Region-based and 
edge-based segmentations are the two major classes of 
segmentation methods.  However, both exhibit a number of 
individual disadvantages.  In this work we present a fully 
probabilistic segmentation framework that integrates Markov 
random fields (MRFs) and deformable models using graphical 
multinets. A graphical model is constructed to represent the 
relationship of the observed image pixels, the region labels 
and the underlying object contour.  We then formulate the 
problem of image segmentation as the one of joint 
region-contour inference and learning in the graphical 
model. Exact inference (segmentation) is intractable and we 
consider a number of approximate tractable solutions. 
Experimental results on numerous difficult medical images 
show that our new hybrid method outperforms both the 
MRF-based and the deformable model-based methods.

Host: Stan Sclaroff