.png){kind=icon}
返回
Gaussian distribution-based facial expression feature extraction network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.patrec.2022.10.015.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
In this paper, we proposed a plug-and-play Gaussian-based Facial Expression Feature Extraction Network (GFE2N) for effective Facial Expression Recognition (FER). We considered facial expression image as a combination of the common information (facial attribute features) across different images and the unique information (facial movement features) that described expression movements. More specifically, GFE2N consisted of two crucial sub-networks: a Gaussian Sampling Network (GSN) and a Movement Feature Generation Network (MFGN). In particular, GSN first simulated latent facial attributes by a Gaussian distri-bution with neutral images. Then, MFGN captured the relationship between latent attributes and overall representations of expression images to generate the sample-specific variations of facial movement fea-tures. The wide ranges of experiments (including on in-the-wild datasets RAF-DB, AffectNet, FERPLUS and in-the-lab dataset CK+) illustrated that the proposed GFE2N can consistently achieve competitive results. GFE2N can also be fused as a subsidiary strategy module to assist improve the existing FER performance. (c) 2022 Published by Elsevier B.V.
Keyword:
Facial expression recognition
Gaussian distribution model
Feature disentanglement
期刊
IF:
3.3
论文数:
8.0K
被引数:
1.6W
机构
暂无机构信息
引用论文
Spontaneous facial micro-expression analysis using Spatiotemporal Completed Local Quantized Patterns.png)
NEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
分享收藏Deep multi-path convolutional neural network joint with salient region attention for facial expression recognitionPATTERN RECOGNITION
PATTERN RECOGNITIONIF7.6
PREAI
分享收藏
查看更多内容