Facial Expression Recognition with Multi-level Integration Disentangled Generative Adversarial Network

Facial Expression Recognition (FER) is a challenging task in computer vision that aims to infer people's emotional states by analyzing and recognizing their facial expressions. However, the extraction of expression features from facial images is usually interfered by other facial features (e.g. pose and appearance), which makes it difficult to improve the performance of the model. To further optimize the model performance, an approach called Multi-level Integration Disentangled Generative Adversarial Network (MID-GAN) is introduced to address the effect of redundant information in facial features on expression features. Compared with traditional expression recognition, MID-GAN makes the following innovations: (1) The model separates irrelevant information from expression features. (2) Inspired by the idea of expression transfer, this method constructs two branches, one for extracting expression features and the other for extracting facial features. (3) Through adversarial training, MID-GAN can transfer the expression from the source domain to the target domain, thereby improving the performance of the model. Experiments on field data sets demonstrate the excellent performance of the proposed method.