A Multi-View Face Expression Recognition Method Based on DenseNet and GAN

Facial expression recognition (FER) techniques can be widely used in human-computer interaction, intelligent robots, monitoring, and other domains. Currently, FER methods based on deep learning have become the mainstream schemes. However, these some problems, such as a large number of parameters, difficulty being applied to embedded processors, fact that accuracy is affected by facial deflection. To solve problem we propose DSC-DenseNet model, which improves standard convolution DenseNet depthwise separable (DSC). wherein face deflection affects effect, posture normalization model GAN: GAN with two local discriminators (LD-GAN) strengthen discriminatory abilities expression-related parts, parts related eyes, eyebrows, mouth, nose. These improve model’s ability retain expressions evidently benefits FER. Quantitative qualitative experimental results Fer2013 KDEF datasets consistently shown superiority our method when working multi-pose images.