Research on Facial Expression Recognition Based on The Combination of Improved LBP and CNN

Abstract

This paper proposes an In view of the problem of insufficient effectiveness of facial expression feature extraction in complex environments, this paper uses CNN to extract global abstract features. Circular LBP is adopted as the basic operator, which fully considers the interaction between the central pixel and the surrounding pixels. Finally, the features extracted by LBP and the features extracted by the deep network are fused in a weighted manner to obtain the final result. The accuracy rate on the Fer2013 data set reaches 74.21%, which has a significant improvement in improving the recognition accuracy compared with other network models