Dictionary learning feature space via sparse representation classification for facial expression recognition

Facial expression recognition (FER) plays a significant role in human-computer interaction. In this paper, adopting a dictionary learning feature space (DLFS) via sparse representation classification (SRC), we propose a method for FER. First, we obtain a difference dictionary (DD) from the feature space by indirectly using an auxiliary neutral training set. Next, we use a dictionary learning algorithm to train the DD; this algorithm considers the samples from the DD are approximately symmetrical structure. Finally, we use SRC to represent and determine the label of each query sample. We then verify out proposed method from the perspective of training samples, dimension reduction methods and Gaussian noise variances using a variety of public databases. In addition, we compare our DLFS_SRC approach with DLFS_CRC and DLFS_LRC approaches on the Extended Cohn-Kanade (CK+) database to analyze recognition results. Our simulation experiments show that our proposed method achieved satisfying performance levels for FER.