论文信息 - MRF parameter estimation by an accelerated method

MRF parameter estimation by an accelerated method

Markov random field (MRF) modelling is a popular method for pattern recognition and computer vision and MRF parameter estimation is of particular importance to MRF modelling. In this paper, a new approach based on Metropolis-Hastings algorithm and gradient method is presented to estimate MRF parameters. With properly chosen proposal distribution for Metropolis-Hastings algorithm, the Markov chain constructed by the method converges to stationary distribution quickly and it gives a good estimation result.

Yihua Yu | QianSheng Cheng

[1] Jun S. Liu,et al. Monte Carlo strategies in scientific computing , 2001 .

[2] J. Besag. Spatial Interaction and the Statistical Analysis of Lattice Systems , 1974 .

[3] Wojciech Pieczynski,et al. Estimation of Generalized Multisensor Hidden Markov Chains and Unsupervised Image Segmentation , 1997, IEEE Trans. Pattern Anal. Mach. Intell..

[4] N. Metropolis,et al. Equation of State Calculations by Fast Computing Machines , 1953, Resonance.

[5] W. K. Hastings,et al. Monte Carlo Sampling Methods Using Markov Chains and Their Applications , 1970 .

[6] Donald Geman,et al. Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images , 1984, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[7] Lei Wang,et al. MRF parameter estimation by MCMC method , 2000, Pattern Recognit..

[8] Emile H. L. Aarts,et al. Simulated Annealing: Theory and Applications , 1987, Mathematics and Its Applications.

[9] Gerhard Winkler,et al. Image analysis, random fields and dynamic Monte Carlo methods: a mathematical introduction , 1995, Applications of mathematics.

[10] Haluk Derin,et al. Modeling and Segmentation of Noisy and Textured Images Using Gibbs Random Fields , 1987, IEEE Transactions on Pattern Analysis and Machine Intelligence.