Adaptive Graph-Based Feature Normalization for Facial Expression Recognition
This work addresses data uncertainties in FER, which is a domain-specific problem for facial expression analysis, and is incremental as it builds on existing methods by incorporating associative relations of expressions.
The paper tackles the problem of data uncertainties in Facial Expression Recognition (FER) caused by ambiguous images and subjective annotations, proposing an Adaptive Graph-based Feature Normalization (AGFN) method that normalizes feature distributions using expression associations, achieving accuracies of 91.84% on FERPlus and 91.11% on RAF-DB, with significant gains of 3.38% and 4.52% over existing works when mislabeled data increases to 20%.
Facial Expression Recognition (FER) suffers from data uncertainties caused by ambiguous facial images and annotators' subjectiveness, resulting in excursive semantic and feature covariate shifting problem. Existing works usually correct mislabeled data by estimating noise distribution, or guide network training with knowledge learned from clean data, neglecting the associative relations of expressions. In this work, we propose an Adaptive Graph-based Feature Normalization (AGFN) method 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 network. Our method outperforms state-of-the-art works with accuracies of 91.84% and 91.11% on the benchmark datasets FERPlus and RAF-DB, respectively, and when the percentage of mislabeled data increases (e.g., to 20%), our network surpasses existing works significantly by 3.38% and 4.52%.