Title: Micro-expression recognition method based on CNN-LSTM hybrid network

Authors: Wang Qingqing

Addresses: Jilin Animation Institute, High Tech Industrial Development Zone, Changchun City, Jilin Province, China

Abstract: Micro-expression has the characteristics of short duration and weak intensity, which makes it difficult to recognise. This paper is mainly to study the recognition of micro-expressions. Combined with the deep learning method, a hybrid neural network model based on CNN-LSTM is proposed, which mainly includes two parts. In one part, convolution neural network is used to extract micro-expression features. In the other part, LSTM correlation model is used to extract the information hidden in the time domain of micro-expression, which makes up for the lack of dynamic features extracted by CNN model. The proposed algorithm is tested on CASMEII data set. The experimental results show that the text accuracy of this algorithm reached 74.8%. It can be seen that the model can make full use of the information of features in the time domain, so as to more effectively carry out micro-expression recognition, and has more advantages than traditional methods.

Keywords: convolutional neural network; long short term memory; micro-expression recognition; deep learning.

DOI: 10.1504/IJWMC.2022.125537

International Journal of Wireless and Mobile Computing, 2022 Vol.23 No.1, pp.67 - 77

Received: 05 Nov 2021
Received in revised form: 16 May 2022
Accepted: 01 Jun 2022

Published online: 13 Sep 2022 *

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