Discriminative attention-augmented feature learning for facial expression recognition in the wild

Facial expression recognition (FER) in-the-wild is challenging due to unconstraint settings such as varying head poses, illumination, and occlusions. In addition, the performance of a FER system significantly degrades due to large intra-class variation and inter-class similarity of facial expressions in real-world scenarios. To mitigate these problems, we propose a novel approach, Discriminative Attention-augmented Feature Learning Convolution Neural Network (DAF-CNN), which learns discriminative expression-related representations for FER. Firstly, we develop a 3D attention mechanism for feature refinement which selectively focuses on attentive channel entries and salient spatial regions of a convolution neural network feature map. Moreover, a deep metric loss termed Triplet-Center (TC) loss is incorporated to further enhance the discriminative power of the deeply-learned features with an expression-similarity constraint. It simultaneously minimizes intra-class distance and maximizes inter-class distance to learn both compact and separate features. Extensive experiments have been conducted on two representative facial expression datasets (FER-2013 and SFEW 2.0) to demonstrate that DAF-CNN effectively captures discriminative feature representations and achieves competitive or even superior FER performance compared to state-of-the-art FER methods.