A structured sparse learning approach for efficient facial feature description

The classical local binary pattern (LBP) method for facial feature description leads to a high feature dimensionality which requires expensive computational cost for face recognition and ignores the difference of contributions by different features in the same region. In this paper, we propose a structured sparse learning approach for efficient facial feature description. Firstly, a structured sparse representation scheme is employed to learn the feature evaluation vector, and then a new facial feature description model is constructed for face recognition. Specially, the proposed approach focuses on selecting the salient regions and features for efficient facial feature description. Experimental results show that the proposed method achieves better performance with lower feature dimensionality.