Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (22): 251-258.DOI: 10.3778/j.issn.1002-8331.2206-0329

• Graphics and Image Processing • Previous Articles     Next Articles

Facial Expression Recognition Method with Attention-Free Capsule Network

XU Xuebin, LIU Chenguang, LU Longbin, CAO Shuxin, XU Zongyu   

  1. 1.School of Computer Science, Xi’an University of Posts & Telecommunications, Xi’an 710121, China
    2.Shannxi Key Laboratory of Network Data Analysis and Intelligent Processing, Xi’an University of Posts & Telecommunications, Xi’an 710121, China
  • Online:2023-11-15 Published:2023-11-15

无注意力胶囊网络的面部表情识别方法

许学斌,刘晨光,路龙宾,曹淑欣,徐宗瑜   

  1. 1.西安邮电大学 计算机学院,西安 710121
    2.西安邮电大学 陕西省网络数据分析与智能处理重点实验室,西安 710121

Abstract: Expression recognition technology can analyze the emotional activities of recognized objects from human expressions. Aiming at the problem that an effective feature extraction and mapping model cannot be established due to the complex spatial relationship and feature information in facial expression images, the spare multilayer perceptron(sMLP) uses a small amount of parameters to allow each spatial location to perform, and the capsule network can also express the spatial pose information of the feature, so this paper proposes a new facial expression recognition model sMLP-CapsNet to improve the ability of expression recognition spatial relationship mapping. The CK+ dataset and the RAF-DB dataset are used to extract facial expression picture features from contour to detail by using an improved capsule neural network to achieve facial expression classification. Compared with other facial expression recognition algorithms, the accuracy of the model is significantly improved, and the recognition rates on the CK+ dataset and RAF-DB dataset can reach 99.48% and 85.69%, respectively, showing the advanced nature of the algorithm.

Key words: deep learning, capsule neural network, facial expression recognition, spare multilayer perceptron(sMLP), sMLP-CapsNet

摘要: 表情识别技术可以从人类的表情中分析出识别对象的情感活动,针对面部表情图像复杂的空间关系和特征信息时,不能建立有效特征提取和映射模型的问题,稀疏多层感知机(spare multilayer perceptron,sMLP)使用很少的参数量让每个空间位置进行交流,而胶囊网络也可以表现特征的空间姿态信息,因此提出了一种新的面部表情识别模型sMLP-CapsNet,以提升表情识别空间关系映射的能力。采用CK+数据集和RAF-DB数据集,通过改进的胶囊神经网络从轮廓到细节提取面部表情图片特征,进而实现面部表情分类。相比于其他面部表情识别算法,模型精度提升效果明显,在CK+数据集和RAF-DB数据集上分别可达到99.48%以及85.69%的识别率,展现了该算法的先进性。

关键词: 深度学习, 胶囊神经网络, 面部表情识别, 稀疏多层感知机(sMLP), sMLP-CapsNet