Learning Discriminative Dictionary for Facial Expression Recognition

Sparse coding is currently an active subject in signal processing, computer vision, pattern recognition, etc. Fisher discrimination dictionary learning (FDDL) is a recently developed discriminative dictionary learning method and exhibits promising performance for classification. However, FDDL could not capture the locality structure of data, and it produces discriminative sparse coding coefficients, which is not effective enough for classification. To address these issues, this paper proposes an advanced version of FDDL by integrating data locality and group Lasso regularization in the procedure of FDDL's sparse coding. The proposed method is used to learn locality- and group-sensitive discriminative dictionary for facial expression recognition. Our experimental results on two public facial expression databases, i.e., the JAFFE database and the Cohn-Kanade database, demonstrate the effectiveness of the proposed method on facial expression recognition tasks, giving a significant performance improvement over FDDL.