Self-Paced Neutral Expression-Disentangled Learning for Facial Expression Recognition

The accuracy of facial expression recognition is typically influenced by the following factors: high similarities across different expressions, disturbing factors, and the subtle and rapid micro-movements of the face. One potential solution to overcome these obstacles is to leverage the hidden neutral information present in neutral expression images. In this paper, we propose a Self-Paced Neutral Expression-Disentangled Learning (SPNDL) model, which aims to disentangle the neutral information from facial expressions, facilitating the extraction of crucial features and variations. This approach enables the capture of discriminative details within similar expressions and the detection of micro-facial movements. To enhance the learning of these neutral expression-disentangled features (NDFs) and mitigate the challenges posed by non-convex optimization, we propose a self-paced learning (SPL) strategy based on NDFs during the training phase. SPL learns samples from easy to complex by increasing the number of samples selected into the training process, which enables to effectively suppress the negative impacts caused by low-quality samples and inconsistently distributed NDFs. Experiments on three popular databases (i.e., CK+, Oulu-CASIA, and RAF-DB) show the effectiveness of our proposed method.