Nonlinear Fusion of Colors to Face Authentication Using EFM Method

Abstract

The authentication systems of face generally used the grayscale face image as input, but in this paper we studied the contribution of the color to the authentication system of face. For the extraction of face characteristics for the data base, we tested different spaces colors on the Enhanced Fisher linear discriminant Model (EFM) which is presented as an alternative features extraction algorithm to Principal Component Analysis (PCA) widely used in automatic face recognition. And once the characteristic vector is extracted, the next stage consists of comparing it with the vector characteristic of face which is authenticated, and with the use of each component color alone at the input of this system, we calculated the error rates in the two sets of validation and test for the data base XM2VTS according to the protocol of Lausanne. Finally, the results obtained in different spaces or components colorimetric are combined by the use of a nonlinear fusion with a simple neuron network MLP (Multi layer perceptron), the results obtained confirm the efficient of color to improve the performance of an authentication system of face.