Abstract | ||
---|---|---|
A key difficulty in the use of Gibbs prior distributions in Bayesian image analysis is the intractability of the normalisation constant. One approach is to perform off-line simulations which allow a calibration of normalisation constant against prior parameter. In this paper the reverse-logistic regression approach to calibration will be examined for various Gibbs distributions and explicit parametric equations will be proposed. A simple method for combining separate calibrations will be illustrated and the relationship between normalisation constant and image size will be explored with an empirical approximation proposed. |
Year | DOI | Venue |
---|---|---|
2002 | 10.1023/A:1020752516117 | Statistics and Computing |
Keywords | Field | DocType |
Bayesian image analysis,Markov random field,Markov chain Monte Carlo,normalising constant,partition function,reverse-logistic regression | Mathematical optimization,Boltzmann distribution,Parametric equation,Markov chain Monte Carlo,Regression,Partition function (statistical mechanics),Markov random field,Statistics,Mathematics,Gibbs sampling,Bayesian probability | Journal |
Volume | Issue | ISSN |
12 | 4 | 1573-1375 |
Citations | PageRank | References |
1 | 0.40 | 2 |
Authors | ||
1 |
Name | Order | Citations | PageRank |
---|---|---|---|
R. G. Aykroyd | 1 | 3 | 2.15 |