Discriminative Deep Feature Learning for Facial Emotion Recognition

Emotion recognition is an important task in facial expression analysis with various potential applications. The goal of this task is to classify facial images into seven classes: disgust, neutral, sad, happy, fear, surprise and angry. In this paper, we propose a discriminative deep feature learning approach with dense convolutional networks (DenseNet) for facial emotion recognition. Particularly, we employ an auxiliary loss, namely center loss, to regulate the training process of neural networks in order to reduce the intra-class variation of the deep features and, hence, to enhance the discriminative power of the learned networks. The experimental results show that our proposed approach achieves superior performance in comparison with other recent state- of-the-art methods on the well-known FERC-2013 dataset.