Micro-expression recognition convolutional network based on dual-stream temporal-domain information interaction

Abstract

There is a shortage of experimental data in the use of deep learning method for micro expression recognition. The neural network can obtain limited knowledge in the learning process and it is difficult to improve the accuracy. To tackle the problems, a dual-stream temporal-domain information interaction neural convolution network (Dual Scale Temporal Interactive Convolution Neural Network, DSTICNN) network is constructed to process micro-expression sequences and realize the automatic recognition of micro-expression. By improving the deep mutual learning strategy to guide the network to learn different temporal domain information of the same image sequence, the final recognition rate can be improved. The algorithm is tested on CASMEII and SMIC databases, and the results show that the algorithm in this paper can effectively improve the recognition rate of micro expressions, and the overall algorithm is better than existing algorithms.