A New Facial Expression Recognition Scheme Based on Parallel Double Channel Convolutional Neural Network

© 2020, Springer Nature Switzerland AG. The conventional deep convolutional neural networks in facial expression recognition are confronted with the training inefficiency due to many layered structure with a large number of parameters, in order to cope with this challenge, in this work, an improved convolutional neural network—with parallel double channels, termed as PDC-CNN, is proposed. Within this model, we can get two different sized feature maps for the input image, and the two different sized feature maps are combined for the final recognition and judgment. In addition, in order to prevent over-fitting, we replace the traditional RuLU activation function with the Maxout model in the fully connected layer to optimize the performance of the network. We have trained and tested the new model on JAFFE dataset. Experimental results show that the proposed method can achieve 83% recognition rate, in comparison with the linear SVM, AlexNet and LeNet-5, the recognition rate of this method is improved by 14%–28%.