Multiview discriminative learning for age-invariant face recognition

In this paper, we propose a new multiview discriminative learning (MDL) method for age-invariant face recognition, which is a challenging and important problem in many practical face recognition systems. Motivated by the fact that local appearance features are more robust to age variations, we first extract three different local feature descriptors including scale invariant feature transform (SIFT), local binary patterns (LBP) and gradient orientation pyramid (GOP) for each face image to exploit the discriminative information. Then, we develop a discriminative learning method with multiview feature representations, called MDL, to project different types of local features into a latent discriminative subspace where the intraclass variation of each feature is minimized, the interclass variation of each feature and the correlation of different features of the same person are maximized, simultaneously, such that more discriminative information can be boosted for recognition. Experimental results on the widely used MORPH and FG-NET face aging datasets are presented to show the efficiency of the proposed approach.