Facial Expression Recognition Based on Sobel Operator and Improved CNN-SVM

Aiming at the problem that traditional convolutional neural networks (CNN) have incomplete facial expression feature extraction and optimizable recognition accuracy in facial expression recognition (FER), this paper proposes a FER model based on improved CNN for Sobel edge detection and fused support vector machine (SVM). In this model, the VGG11 network is used to realize the feature extraction of facial expressions, and the L2-SVM is used to replace the softmax activation function to realize facial expression classification. This paper has been verified on the CK+ and JAFFE data sets, and the experimental results show that on the CK+ data set, after replacing the softmax activation function with L2-SVM, the accuracy rate has been improved by 3.02%. On this basis, in the image preprocessing stage, after adding Sobel edge detection, the accuracy has been improved by 3.71%. On the JAFFE data set, after using L2-SVM, the accuracy rate has increased by 2.16%, and after adding Sobel edge detection, it has increased by 2.17%.