A discriminative deep association learning for facial expression recognition

Xing Jin, Wenyun Sun, Zhong Jin

Abstract: Deep learning based facial expression recognition becomes more successful in many applications. However, the lack of labeled data is still a bottleneck for better recognition performance. Thus, it is of practical significance to exploit the rich unlabeled data for training deep neural networks (DNNs). In this paper, we propose a novel discriminative deep association learning (DDAL) framework. The unlabeled data is provided to train the DNNs with the labeled data simultaneously, in a multi-loss deep network based on association learning. Moreover, the discrimination loss is also utilized to ensure intra-class clustering and inter-class centers separating. Furthermore, a large synthetic facial expression dataset is generated and used as unlabeled data. By exploiting association learning mechanism on two facial expression datasets, competitive results are obtained. By utilizing synthetic data, the performance is increased clearly.