ResNet with Attention Mechanism and Deformable Convolution for Facial Expression Recognition

Due to the influence of face occlusion, side face and other factors, the recognition accuracy still needs to be improved in facial expression recognition(FER). This paper proposes a deep residual network incorporating an attention mechanism and deformable convolution to improve this phenomenon. The improved network is an attention mechanism embedded in the residual blocks with position information and a deformable convolution is added at the input to enhance the feature extraction capability for different locations of the image. This paper has been proved on the RAF-DB and FERPlus datasets, and the experimental results show that the accuracy of the improved network is 1.76% higher than the original network on the RAF-DB dataset and 2.76% higher than the original network on the FERPlus dataset.