Learning parameters of Gibbs random fields using unconditional and conditional MLE of potentials

  • Georgy Gimel'farb
Poster Papers
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1451)


Two parameter learning schemes for Gibbs random field image models with translation invariant multiple pairwise pixel interactions are discussed. The schemes allow to estimate both the interaction structure and strengths (Gibbs potentials) from a given training sample. The first scheme is based on the unconditional MLE of the potentials. The estimates are specified in an implicit form and can be obtained in three steps: (i) an analytic first approximation of the potentials for a big many possible neighbours, (ii) a search for most characteristic neighbours, and (iii) a stochastic approximation refinement of the estimates for a chosen set of neighbours. The second scheme uses the conditional MLE suggesting that the training sample has the least upper bound (top rank) in its total Gibbs energy within the parent population. This scheme allows to deduce an explicit, to scaling factors, analytic form of the potentials. Then only the scaling factors have to be learnt using their MLE in a like three-step manner. The conditional MLE of the potentials seems to be close to the unconditional ones and extends capabilities of the Gibbs image models.


  1. 1.
    Barndorrf-Nielsen, O.: Information and Exponential Families in Statistical Theory. Wiley (1978)Google Scholar
  2. 2.
    Besag, J. E.: Spatial interaction and the statistical analysis of lattice systems”. J. Royal Statistical Soc. B36 (1974) 192–236Google Scholar
  3. 3.
    Chellappa, R., Jain, A. (Eds): Markov Random Fields. Theory and Application. Academic Press (1993)Google Scholar
  4. 4.
    Dobrushin, R. L., Pigorov, S. A.: Theory of random fields. Proc. 1975 IEEE-USSR Joint Workshop on Information Theory, December 15–19, 1975, Moscow, USSR. IEEE (1976) 39–49Google Scholar
  5. 5.
    Gimel'farb, G. L.: Analytic approximation of Gibbs potentials to model stochastic textures. Proc. First Joint Australia & New Zealand Biennial Conf. on Digital Image & Vision Computing: Techniques and Applications (DICTA'97/IVCNZ'97), December 10–12, 1997, Albany, Auckland, New Zealand. Massey University (1997) 153–158Google Scholar
  6. 6.
    Gimel'farb, G. L.: Texture modeling by multiple pairwise pixel interactions. IEEE Trans. Pattern Anal. Machine Intell. 18 (1996) 1110–1114CrossRefGoogle Scholar
  7. 7.
    Gimel'farb, G. L.: Gibbs models for Bayesian simulation and segmentation of piecewise-uniform textures. Proc. 13th IAPR Int. Conf. on Pattern Recognition, August 25–29, 1996, Vienna, Austria. TUWien (1996) B 591–595Google Scholar
  8. 8.
    Gimel'farb, G. L.: Non-Markov Gibbs texture model with multiple pairwise pixel interactions. Ibid, 760–764Google Scholar
  9. 9.
    Gimel'farb, G. L., Jain, A. K.: On retrieving textured images from an image database. Pattern Recognition 29 (1996) 1461–1483CrossRefGoogle Scholar
  10. 10.
    Winkler, G.: Image Analysis, Random Fields, and Dynamic Monte Carlo Methods. Springer (1995)Google Scholar
  11. 11.
    Younes, L. Estimation and annealing for Gibbsian fields. Annales de l'Institut Henri Poincare. 24 (1988) 269–294Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1998

Authors and Affiliations

  • Georgy Gimel'farb
    • 1
  1. 1.CITR, Department of Computer Science, Tamaki CampusThe University of AucklandAucklandNew Zealand

Personalised recommendations