Expression Recognition of Dual Channels Model System Based on Mini_Xception_SE
Chan Su, Jianguo Wei · 2020
As an important attribute of human face, expression is widely used in human-computer interaction, artificial intelligence and other fields. As an important biometric recognition technology, expression recognition has become a hot research topic in the field of computer vision. With the continuous development of deep learning, appropriate neural network and a large number of preprocessed data can train a high-performance model, so as to improve the accuracy and convenience of expression recognition. Because the local details of the face determine the expression information, this paper not only uses the face detection technology in the image preprocessing, but also uses Difference of Gaussian pyramid to obtain the detailed information of the facial expression image to enhance the image detail features. Because of the large amount of parameters in Xception model, this paper proposes a dual channels model based on mini_Xception_SE for facial expression classification. The experimental results on fer2013 and CK+ dataset show that the feature information of facial expressions in different facial regions can be effectively extracted by applying detail enhanced image to the dual channels model based on Mini Xception SE, so as to improve the accuracy of facial expression recognition.