Adaptive graph-based feature normalization for facial expression recognition

Facial Expression Recognition (FER) suffers from data uncertainties caused by ambiguous facial images and annotators’ subjectiveness, resulting in excursive semantic and feature covariate shifting problems. Existing works usually correct mislabeled data by estimating noise distribution, or guide network training with knowledge domain that learned from clean data, neglecting the associative relations of expression samples. In this work, we propose an Adaptive Graph-based Feature Normalization (AGFN) to protect FER models from data uncertainties by normalizing feature distributions with the association of expressions. Specifically, we propose a Poisson graph generator to adaptively construct topological graphs for samples in each mini-batches via a sampling process, and correspondingly design a coordinate descent strategy to optimize proposed model. Our method outperforms state-of-the-art works with accuracies of 91. 84%, 91. 11% and 61. 38% on three benchmark datasets, i. e. , FERPlus, RAF-DB and AffectNet. Especially, when the percentage of mislabeled data significantly increases (e. g. , to 20%), our method surpasses existing works by 14. 09%, 21. 12% and 13. 67% on above datasets. Our code is publicly available in https: //github. com/X-Lab-CN/AGBFN.

Authors

• Yu-Jie Xiong
• Qingqing Wang
• Yangtao Du
• Yue Lu School of Communication and Electronic Eng. , East China Normal University

Keywords

• Facial expression recognition
• Data uncertainties
• Feature normalization
• Poisson graph generator