Facial Expression Recognition via a CBAM Embedded Network

Abstract

Abstract Facial expression recognition plays an important role in face recognition. Face recognition helps computer to figure out people faces from a simple scenery and to recognize who they are. But facial expression recognition assists computer to analyze the emotion state of one single person in order to improve the human-computer interaction experience. For facial expressions, there are certainly lots of conspicuous features for observers to look on, like shapes of eyes, mouth. When people smile, their lips turn up and their eyebrows bend downward. In this way, we get features of smile. And in same way, we can also do the same to anger, sadness, and surprise etc. Single neural network can make it. However, the accuracy is always low, nearly 70% up and down, which always companied with fluctuation, like electrocardiogram, though not that severe, according to experiments. Therefore, we choose to include Convolutional Block Attention Module (CBAM) into some layers of VGG network we used to improve the accuracy and also the stability. CBAM is an effective attention module for neural networks to concentrate on features. With CBAM, we simplify the scenery that networks need to analyze, and then we can finally make an improvement. This way, we can build better human-computer interfaces.