New approaches to parameter estimation in Gibbs pattern recognition models

The paper considers approach to the image recognition problem that is based on random Gibbs fields. Estimation of parameters of Gibbs models. One of the possible approach to this problem is maximum entropy principle. According to this principle we should seek such distribution which is as non-informative as available prior information. the paper suggests another approach, which seeks such distributions, for which maximum and minimum values of probabilities of random object is attained.