Training of the Beta wavelet networks by the frames theory: Application to face recognition

Abstract

A wavelets neural network is a hybrid classifier composed of a neuronal contraption and wavelets as functions of activation. Our approach of face recognition is divided in two parts: the training phase and the recognition phase. The first consists in optimizing a wavelets neural network for every training picture face. A new technique of training of these wavelets networks which based on the frames theory is proposed as a remedy to the inconveniences of the classical training algorithms. The specificity of a BWNN to a face and the notion of SuperWavelet have been exploited to propose an approach of face recognition. Finally, we have compared our method of recognition to other ones which are used for face recognition that are applied on the AT&T (ORL) and FERET faces basis. We reached a face recognition rate that exceeds 90% for two images per person in the training step.