Facial expression recognition based on salient patch selection

摘要

Expressions are commonly presented through the motions of different facial regions, thus the selection of discriminative features from prominent regions is crucial to the expression recognition. This paper proposes a novel method for facial expression recognition by exploring the most salient regions for each expression. The main contribution of this paper is using the complete feature set of expressions to replace the whole facial feature, which not only improves the accuracy but also saves time. In order to evaluate the saliency of different facial regions and features, the group lasso scheme is applied. Experiments on Cohn-Kanade (CK+) database show the effectiveness of the proposed method.