Bayes classifier and loss functions

Abstract

This paper deals with the classification of objects into the limited number of classes. The objects are characterized by n-features. The paper focuses on the Bayes classifier based on the probability principle with the fixed number of the features during the classification process. The Bayes classifier which uses criterion of the minimum error was applied on the set of the multispectral data. They represent real images of the Earth surface obtained from remote Earth sensing. The paper describes experience and results obtained during the classification of extensive set of these multispectral data with the Bayes classifier using the symmetric and diagonal loss function.