Proceedings of the 2018 International Conference on Computer Science, Electronics and Communication Engineering (CSECE 2018)

Multi-Network Fusion Based on CNN for Facial Expression Recognition

Authors
Chao Li, Ning Ma, Yalin Deng
Corresponding Author
Chao Li
Available Online February 2018.
DOI
https://doi.org/10.2991/csece-18.2018.35How to use a DOI?
Keywords
facial expression recognition; fusion; multi-network; CNN; SVM
Abstract
We propose a method which is multi-network fusion (MNF) based on CNN to recognize facial expressions. Our experimental data adopts the ICML2013 facial expression recognition contest's dataset (FER-2013) and JAFFE dataset. Based on the classic Tang's network structure and Caffe-ImageNet structure, we perform pre-training separately to extract the optimal initialization parameters which are applied for the MNF. We adjust the MNF's parameters through fine-tuning and use L2-SVM for classification. Our experiment has achieved a high accuracy, and the result shows that the effect of the MNF is more obvious than a single network on the facial expression recognition. In this paper, we will describe the specific MNF structure and our training process, as well as the accuracy on the test set.
Open Access
This is an open access article distributed under the CC BY-NC license.

Download article (PDF)

Proceedings
2018 International Conference on Computer Science, Electronics and Communication Engineering (CSECE 2018)
Part of series
Advances in Computer Science Research
Publication Date
February 2018
ISBN
978-94-6252-487-3
ISSN
2352-538X
DOI
https://doi.org/10.2991/csece-18.2018.35How to use a DOI?
Open Access
This is an open access article distributed under the CC BY-NC license.

Cite this article

TY  - CONF
AU  - Chao Li
AU  - Ning Ma
AU  - Yalin Deng
PY  - 2018/02
DA  - 2018/02
TI  - Multi-Network Fusion Based on CNN for Facial Expression Recognition
BT  - 2018 International Conference on Computer Science, Electronics and Communication Engineering (CSECE 2018)
PB  - Atlantis Press
SN  - 2352-538X
UR  - https://doi.org/10.2991/csece-18.2018.35
DO  - https://doi.org/10.2991/csece-18.2018.35
ID  - Li2018/02
ER  -