Array of Multilayer Perceptrons with No-class Resampling Training for Face Recognition

Abstract

A face recognition (FR) problem involves the face detection, representation and classification steps. Once a face is located in an image, it has to be represented through a feature extraction process, for later performing a proper face classication task. The most widely used approach for feature extraction is the eigenfaces method, where an eigenspace is established from the image training samples using principal components analysis.In the classification phase, an input face is projected to the obtained eigenspace and classified by an appropriate classifier. Neural network classifiers based on multilayer perceptron models have proven to be well suited to this task. This paper presents an array of multilayer perceptron neural networks trained with a novel no-class resampling strategy which takes into account the balance problem between class and no-class examples andincreases the generalization capabilities. The proposed model is compared against a classical multilayer perceptron classifier for face recognition over the AT&T database of faces, obtaining results that show an improvement over the classification rates of a classical classifier.