Design of Facial Expression Recognition Algorithm Based on CNN Model
Yue Luo, Jiaxin Wu, Zhuhao Zhang, Huaju Zhao, Zhong Shu
Abstract
This paper proposes an improved CNN model with the main process of face image preprocessing, face image feature extraction, input test sample training, acquisition of test sample features, face image feature classification, restoration of face expression images, and recognition results. In the convolution activation function, the cumulative sum operation is performed on three aspects of facial expression image pixels, the number of convolution operations, and the number of pooling operations, and a highly integrated facial expression image feature extraction is realized. It is merged with the pooling operation. In the classification loss function, two processes of convolution and pooling operations are considered, and the weight and bias updates in the convolution, pooling, and classification loss functions are included. Relevant experiments and experimental statistics show that the classification accuracy of the facial image expression feature extraction algorithm proposed in this paper is about 87%, and the recognition accuracy of the improved CNN facial expression recognition model proposed in this paper is about 88%.