Metropolis vs Kawasaki Dynamic for Image Segmentation Based on Gibbs Models

Authors:
Xavier Descombes;Eugène Pechersky
Affiliations:
-;-
Venue:
EMMCVPR '99 Proceedings of the Second International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition
Year:
1999

Citing 3
Cited 0

Modeling and Segmentation of Noisy and Textured Images Using Gibbs Random Fields

IEEE Transactions on Pattern Analysis and Machine Intelligence
Simultaneous Parameter Estimation and Segmentation of Gibbs Random Fields Using Simulated Annealing

IEEE Transactions on Pattern Analysis and Machine Intelligence
Numerical recipes in C (2nd ed.): the art of scientific computing

Numerical recipes in C (2nd ed.): the art of scientific computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper we investigate several dynamics to optimize a posterior distribution defined to solve segmentation problems. We first consider the Metropolis and the Kawasaki dynamics. We also compare the associated Bayesian cost functions. The Kawasaki dynamic appears to provide better results but requires the exact values of the class ratios. Therefore, we define alternative dynamics which conserve the properties of the Kawasaki dynamic and require only an estimation of the class ratios. We show on synthetic data that these new dynamics can improve the segmentation results by incorporating some information on the class ratios. Results are compared using a Potts model as prior distribution.