Searching for just a few words should be enough to get started. If you need to make more complex queries, use the tips below to guide you.
Article type: Research Article
Authors: Cen, Shixina | Yu, Yangb | Yan, Gangb | Yu, Minga; b; * | Kong, Yanleib
Affiliations: [a] School of Electronic and Information Engineering, Hebei University of Technology, Tianjin, P.R. China | [b] School of Artificial Intelligence, Hebei University of Technology, Tianjin, P.R. China
Correspondence: [*] Corresponding author. Ming Yu, School of Electronic and Information Engineering, Hebei University of Technology, Xiping Road No. 5340, Beichen District, Tianjin, 300401, P.R. China. E-mail: [email protected].
Abstract: As a spontaneous facial expression, micro-expression reveals the psychological responses of human beings. However, micro-expression recognition (MER) is highly susceptible to noise interference due to the short existing time and low-intensity of facial actions. Research on facial action coding systems explores the correlation between emotional states and facial actions, which provides more discriminative features. Therefore, based on the exploration of correlation information, the goal of our work is to propose a spatiotemporal network that is robust to low-intensity muscle movements for the MER task. Firstly, a multi-scale weighted module is proposed to encode the spatial global context, which is obtained by merging features of different resolutions preserved from the backbone network. Secondly, we propose a multi-task-based facial action learning module using the constraints of the correlation between muscle movement and micro-expressions to encode local action features. Besides, a clustering constraint term is introduced to restrict the feature distribution of similar actions to improve categories’ separability in feature space. Finally, the global context and local action features are stacked as high-quality spatial descriptions to predict micro-expressions by passing through the Convolutional Long Short-Term Memory (ConvLSTM) network. The proposed method is proved to outperform other mainstream methods through comparative experiments on the SMIC, CASME-I, and CASME-II datasets.
Keywords: Micro-expression recognition, multi-scale weighted module, facial action learning module, spatiotemporal network
DOI: 10.3233/JIFS-202962
Journal: Journal of Intelligent & Fuzzy Systems, vol. 41, no. 2, pp. 2905-2921, 2021
IOS Press, Inc.
6751 Tepper Drive
Clifton, VA 20124
USA
Tel: +1 703 830 6300
Fax: +1 703 830 2300
[email protected]
For editorial issues, like the status of your submitted paper or proposals, write to [email protected]
IOS Press
Nieuwe Hemweg 6B
1013 BG Amsterdam
The Netherlands
Tel: +31 20 688 3355
Fax: +31 20 687 0091
[email protected]
For editorial issues, permissions, book requests, submissions and proceedings, contact the Amsterdam office [email protected]
Inspirees International (China Office)
Ciyunsi Beili 207(CapitaLand), Bld 1, 7-901
100025, Beijing
China
Free service line: 400 661 8717
Fax: +86 10 8446 7947
[email protected]
For editorial issues, like the status of your submitted paper or proposals, write to [email protected]
如果您在出版方面需要帮助或有任何建, 件至: [email protected]