Discrete Optimization Methods in Computer Vision

Instructors: Yuri Boykov, Phil Torr, Ramin Zabih

Duration: 3.5 hours

Summary:

Computer vision is full of optimization problems, ranging from low-level vision problems like stereo to high-level problems like object recognition. Most researchers rely primarily on continuous methods to solve these optimization problems. While discrete optimization methods are less common in vision, they can be extremely powerful.

Topics:

Basic concepts: running time, NP hardness, reducibility
Discrete optimization techniques: dynamic programming, graph algorithms
Sampling methods: MCMC, loopy belief propagation
General-purpose techniques ("weak methods"): simulated annealing, genetic algorithms
Example problems: stereo, motion, structure from motion, matching

On the instructors:

Yuri Boykov is an Assistant Professor of Computer Science at the University of Western Ontario. Phil Torr is a Reader in the Department of Computing at Oxford Brookes University. Ramin Zabih is an Associate Professor of Computer Science at Cornell University.

Additional materials:

[new] http://wwwcms.brookes.ac.uk/~philiptorr/eccv_tutorial_2004.htm