Abstract: Firstly, aiming at the problem of facial expression recognition, we proposed a new multi-scale feature selection network recognition method. The network fully combined the advantages of multi-scale network structure and feature selection structure, which could extract the spatial information in the facial static images more effectively. Secondly, in order to verify the recognition performance and generalization ability of the proposed multi-scale feature selection network, we carried out comparison and cross validation experiments with some common methods on two classical facial expression recognition databases. The experiment results show that the proposed network achieves better recognition effect and has good generalization ability, it can be flexibly embedded into the facial expression recognition and analysis system.

Key words: convolutional neural network, facial expression recognition, feature selection mechanism, multi-scale network