A discriminant nonnegative tensor factorization method based on sparse representation classifier

In this paper, a novel discriminant nonnegative tensor factorization method based on sparse representation classifier is proposed for facial expression recognition. It is derived from the nonnegative tensor factorization (NTF) algorithm, and it adopts a discriminant constraint in the objective function. The constraint considers the spatial neighborhood structure and the class information, which is based on the graph embedding theory. Using the discriminant constraint, the obtained parts-based representations would vary smoothly along the geodesics of the data manifold. Finally, the sparse representations are extracted for classification. Experiments are conducted on the JAFFE database and the Cohn-Kanade database. The results demonstrate that our method provides good facial representations and achieves better recognition performance compared with the conventional algorithms.