Title: Dynamic emotion recognition of human face based on convolutional neural network

Authors: Lanbo Xu

Addresses: School of Computer Science and Engineering, Northeastern University, Shenyang, 110167, China

Abstract: To improve the accuracy and speed of facial dynamic emotion recognition, a dynamic emotion recognition of human face based on convolutional neural network is designed. Firstly, locate the facial region, and after greying out the facial region image, use the chaotic frog jump algorithm to enhance the clarity of image features through enhanced processing. Then, analyse the geometric and texture features of the facial image separately to determine key feature points. Finally, after training the convolutional neural network, input geometric features and texture features, calculate the basic parameters, dynamic emotional parameters and feature Loss function of facial dynamic emotional features, and then match geometric features, texture features and emotional template categories to get the final recognition results. Experiment shows that after applying this method, its recognition accuracy is between 97.6%-98.7%, and the maximum recognition time is only 112 ms, indicating that this method has high recognition accuracy and speed.

Keywords: dynamic facial images; emotional recognition; regional positioning; greyscale processing; enhanced processing; feature extraction; convolutional neural network.

DOI: 10.1504/IJBM.2024.140785

International Journal of Biometrics, 2024 Vol.16 No.5, pp.533 - 551

Received: 25 Jul 2023
Accepted: 19 Oct 2023
Published online: 02 Sep 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article