A sparse representation approach for local feature based expression recognition

A novel facial expression recognition method based on sparse representation (SR) is proposed. To enhance the effect of important face region, fisher separation criterion is introduced to calculate the weight of local binary patterns (LBP) patches. Expression recognition technique using the new mathematical theory from sparse representation an compressive sensing is proposed, and the improvements in performance are discussed. Furthermore, a multi-layer sparse representation (MLSR) algorithm is proposed for multi-intensity expression recognition. Experiments showed a better result than conventional sparse representation. More important, MLSR can be generalized to similar classification problems. To verify the efficiency of the proposed methods, a serious of experiments on publicly available databases is performed, including comparisons among SVM, SR, and MLSR. Especially, our method shows better performance against low-resolution images.