Hybrid client specific discriminant analysis and its application to face verification

Abstract

Techniques to perform dimensionality reduction for high dimensional data can vary considerably from each other, leading to different effects on face verification. To address the problem, we introduce a framework called the hybrid client specific discriminant analysis that incorporates various dimensionality reduction methods with client specific subspace. In contrast to the common multidimensional representation, like PCA and LDA, client specific subspace could better describe the diversity of the different faces and has more robust discriminatory information. Moreover, it provides two measures for authentication: a distance to the client template and a distance to the mean of imposter. These two decision scores are combined to achieve significant performance gains. Extensive experiments obtained on the facial databases XM2VTS show the effectiveness of the hybrid client specific discriminant analysis. © Springer-Verlag Berlin Heidelberg 2013.