Contextual classifier combination by Markov random fields-Bayes formalism and evolutionary programming. Application: image classification enhancement

Summary form only given. By using the Markov random fields (MRF)-Bayes formalism with the evolutionary programming, a method of contextual classifier combination for image classification enhancement is proposed. It combines the outputs of the classifiers without taking account of their internal characteristics. So, it can be used for combining any type of classifier. It does not use the training process and the problem of extracting the sufficient and reliable training samples for an accurate estimation of class parameters is avoided. The evolutionary computation is used to process the complementarities and the conflicts that exist between the classifiers. By this way, the actual number of classes is detected. Based on some objective quantitative measures of evaluation and comparison, the proposed method overcomes the disadvantages of the supervised classification methods developed in the literature. Its efficiency can be seen from the experimental results. It gives better results in comparison with those of the classifiers considered separately.