Facial Expression Recognition Based on Convolutional Neural Network
Yue Zhou, Feng Yanyan, Zeng Shangyou, Pan Bing · 2019
Facial expression recognition is an important field of pattern recognition research. Traditional machine learning methods extract features manually. It has insufficient generalization ability and poor stability. Moreover, its accuracy is difficult to improve. In order to achieve better facial expression recognition, this paper designs a modular multi-channel deep convolutional neural network. To avoid overfitting, the network output uses a global average layer. Data enhancement on the dataset before training can improve the generalization ability of the model. Test the performance of network on the FER2013 emoticon dataset. The accuracy of expression recognition is 68.4%. It performs a prediction for about 0.12s. Compared to other recognition algorithms, network has certain advantages. Finally, a real-time facial expression recognition system is constructed by using the trained recognition model. The experimental results show that the system can effectively recognize facial expressions in real time.