Face Expression Recognition Based on Convolutional Neural Network

In order to reduce the complexity for extracting artificial features from the face image in facial expression recognition (FER), a novel method is proposed based on convolutional neural network (CNN) in this paper. This method first preprocesses the facial expression images, then some trainable convolution kernels are used to extract facial expression features, and second, the largest pooling layer is used to fewer dimensions, finally seven types of facial expressions are recognized with the Softmax classifier. The proposed method is verified with Kaggle facial expression recognition challenge dataset (FER2013). The experimental results show that the method has good recognition performance and generalization ability.