Problem of optimal pertinent parameter selection in buried conductive tag recognition

Abstract

The structural pattern recognition passes by an extraction stage of a certain number of characteristic features of the form. In the majority of the cases, it is not necessary to use the whole of the primitives extracted to obtain good performances of the recognition system. One uses the criteria of feature selection like Fisher or the criteria based on covariance matrix to determine the optimal primitives which characterize the best the form. We will show that these criteria give the most discriminating primitives but not necessarily the most optimal for a given classifier.