Face recognition using Multi-Layered Perceptrons

which also reject never seen faces ?

Paolo Frasconi Marco Gori

Giovanni Soda

Dipartimento di Sistemi e Informatica, Universit?a di Firenze

Via di Santa Marta 3 - 50139 Firenze - Italy

Tel. (39) 55-4796265, Fax (39) 55-4796363

e-mail : frasconi, marco @ingfi1.cineca.it

Alberto De Curtis Fabrizio Innocenti

Officine Galileo - Firenze

Abstract

In this paper we show how can Multi-Layered Networks be used for face recognition. Preliminary experiments have demonstrated that these networks can indeed discriminate among different faces, but that they fail in rejecting never seen faces. We suggest a simple architectural solution capable of solving this problem and keeping the already shown skill of discriminating faces.
Index Terms - Back Propagation, Face Recognition.

I Introduction

Recognizing faces is not only an amazing and stimulating pattern recognition exercise, but also an important practical problem for cases where an automatic security check is required. This problem has been recently investigated using neural networks [2, 1]. Some researchers have approached the problem using Backpropagation with neural classifiers

?This research was partially supported by MURST 40%, and Officine Galileo.