Local discriminant time-frequency atoms for signal classification

Abstract

Three methods to select discriminant time-frequency atoms from the Gabor time-frequency dictionary are proposed. The first method performs a discriminant pursuit in its selection, the second method leads to atoms with the most discriminant power, and the third method combines the first two methods. The time-frequency atoms selected by the methods extract discriminant features among different classes of signals. Experimental results on the classification of simulated data sets (triangular waveforms) and real data sets (speech signals) using the extracted features are presented. © 1999 Elsevier Science B.V. All rights reserved.