融合深度学习和稀疏表示的人脸表情识别方法研究  

A Research of Facial Expression Recognition Method Based on Deep Learning and Sparse Representation

作　　者：沈利迪[1,2] SHEN Li-di(Taizhou Vocational and Technical College,College of Information Technology,Taizhou 318000 China;Zhejiang University of Technology,College Of Computer Science&Technology,Hangzhou 310000 China)

机构地区：[1]台州职业技术学院信息技术学院 [2]浙江工业大学计算机科学与技术学院

出　　处：《自动化技术与应用》2019年第10期93-97,共5页Techniques of Automation and Applications

基　　金：高等学校访问学者专业发展项目(编号FX2016122);台州职业技术学院校级重点课题(编号2017ZD03)

摘　　要：随着人机交互和机器学习技术的迅速发展,人脸表情的识别技术受到越来越多研究者的关注。为提高人脸表情的识别率,本文提出了将深度学习和稀疏表示分类方法相结合的图像识别算法。本算法先是利用深度卷积神经网络构建提取人脸表情特征的网络模型,然后利用该网络模型提取样本的特征向量,并建立表情的超完备字典,最后利用稀疏表示分类算法进行人脸表情识别。实验在LFW和AR两个数据集上进行测试,结果证明本文提出的算法有较好的识别效果。With the rapid development of human-computer interaction and machine learning technology,the recognition technology of facial expressions attracts more and more researchers'attention.In order to improve the recognition rate of facial expressions,this paper proposes an algorithm combining deep learning with sparse representation classification.Firstly,in the algorithm,deep learning convolutional neural network is taken to construct a network model for extracting facial expression features.Then,the network model is used to extract the feature vectors of the samples,andestablish the super-complete dictionary of expressions.Finally,the sparse representation classification algorithm is used to perform facial expressionsidentification.The experiment is eimplemented on data sets of LFW and AR.The results show that the proposed algorithm has better recognition effect.

关 键 词：深度学习 卷积网络 特征提取 稀疏表示 表情识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]