A simple upper bound to the Bayes error probability for feature selection

In this paper, feature selection in multiclass cases for classification of remote-sensing images is addressed. A criterion based on a simple upper bound to the error probability of the Bayes classifier for the minimum error is proposed. This criterion has the advantage of selecting features having a link with the error probability with a low computational load. Experiments have been carried out in order to compare the performances provided by the proposed criterion with the ones of some of the widely used feature-selection criteria presented in the remote-sensing literature. These experiments confirm the effectiveness of the proposed criterion, which performs slightly better than all the others considered in the paper.