Sparse local fisher discriminant analysis for facial image analysis

In this paper, we propose a novel feature extraction method called sparse local Fisher discriminant analysis (SLFDA), which is an extension of the local Fisher discriminant analysis (LFDA) algorithm. The proposed method projects the training samples into the range space of local total scatter matrix. Then, it gives the explicit characterization for all solutions of the LFDA. To obtain the sparse projection vectors, we try to find the solution with minimum ℓ₁-norm from all minimum dimensional solutions of the LFDA. This problem is usually formulated as a ℓ₁-minimization problem and is solved by accelerated linear Bregman method. The convergence is an extension of the original accelerated linear Bregman method and is also given in this paper. Experiments results on face and facial expression recognition are presented to demonstrate the effectiveness of the proposed method.