IVSA: Facial Expression Recognition Method with Salient Attention

Recently, facial expression recognition (FER) has become an important topic in computer vision research. With the advance of artificial intelligence, the performance of model about FER has made great progress and improvement. To further enhance the ability of extracting significant features and enhancing the robustness of the model, we present a innovative facial expression recognition framework based on convolutional neural network and attention module. Concretely, we add the L2 norm features in CBAM and re-scale the channel weights. Salient attention block is used to suppress the insignificant feature and enhance the weight of salient features, which improves the performance and robustness of the model. Finally, without using extra training data, IVSA achieves the highest single-model accuracy of 72.44%, which is improved by 1.14% compared with the previous methods. Extensive experiments prove the effectiveness of the model and framework.