Sparse deep feature learning for facial expression recognition. (December 2019)

• Record Type: Journal Article
• Authors: Xie, Weicheng; Jia, Xi; Shen, Linlin; Yang, Meng
• Is Part Of: Pattern recognition. Volume 96(2019:Dec.)
• Publication Date: 2019-12
• Subjects: Expression recognition -- Feature sparseness -- Deep metric learning -- Fine tuning -- Generalization capability; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2019.106966
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A new framework of facial expression recognition is proposed, where different feature sparseness strategies are embedded in deep networks and further investigated. Feature sparseness of the fully-connected layer input is embedded into a deep network to boost the feature generalization ability, which includes less parameters, while achieves better performance than other network sparseness. The advantage of the feature sparseness is evaluated with a toy model based on a quantitative metric. The deep metric learning achieved competitive recognition rates and state-of-the-art cross-database performance on four benchmark expression databases, i.e. FER2013, CK+, Oulu-CASIA and MMI. Abstract: While weight sparseness-based regularization has been used to learn better deep features for image recognition problems, it introduced a large number of variables for optimization and can easily converge to a local optimum. The L2-norm regularization proposed for face recognition reduces the impact of the noisy information, while expression information is also suppressed during the regularization. A feature sparseness-based regularization that learns deep features with better generalization capability is proposed in this paper. The regularization is integrated into the loss function and optimized with a deep metric learning framework. Through a toy example, it is showed that a simple network with the proposed sparseness outperforms the one with the L2-norm regularization. Furthermore, the proposed approach achieved competitive performances on four publicly available datasets, i.e., FER2013, CK+, Oulu-CASIA and MMI. The state-of-the-art cross-database performances also justify the generalization capability of the proposed approach.