Facial Expression Recognition System Based on Deep Residual Fusion Neural Network

Rich and varied facial expressions are the intuitive carriers for transmitting emotional information to each other. Due to the variety of facial expressions, the extraction of features is quite difficult. The traditional manual extraction method can neither achieve better recognition accuracy nor guarantee the recognition efficiency. This paper uses 18-layer residual neural network, and realizes permanent mapping by means of the short-circuit connection of residual modules to ensure the network capability of deep structures. At the same time, the CLBP texture features are extracted, and the two are innovatively combined to form a more representative description feature. The experimental results show that compared with the DCNN, DBN and other networks, the convergence time is shorter and the average recognition rate is 93.24%, which is nearly 5% higher.