电子学报 ›› 2018, Vol. 46 ›› Issue (7): 1710-1718.DOI: 10.3969/j.issn.0372-2112.2018.07.024

• 学术论文 • 上一篇    下一篇

基于稀疏特征挑选和概率线性判别分析的表情识别研究

张瑞, 蒋晨之, 苏剑波   

  1. 上海交通大学自动化系系统控制与信息处理教育部重点实验室, 上海 200240
  • 收稿日期:2017-03-21 修回日期:2018-03-04 出版日期:2018-07-25
    • 通讯作者:
    • 苏剑波
    • 作者简介:
    • 张瑞,男.1991年9月生于安徽.上海交通大学自动化系硕士研究生.研究方向为人脸识别、表情识别.E-mail:Zander_Ray@sjtu.edu.cn
    • 基金资助:
    • 国家自然科学基金 (No.61533012,No.91748120,No.61521063)

Expression Recognition Based on Sparse Selection and PLDA

ZHANG Rui, JIANG Chen-zhi, SU Jian-bo   

  1. Ministry of Education Key Laboratory of System Control and Information Processing, Department of Automation, Shanghai Jiao Tong University, Shanghai 200240, China
  • Received:2017-03-21 Revised:2018-03-04 Online:2018-07-25 Published:2018-07-25
    • Corresponding author:
    • SU Jian-bo
    • Supported by:
    • National Natural Science Foundation of China (No.61533012, No.91748120, No.61521063)

摘要: 提出一种基于稀疏特征挑选(Sparse selection)和概率线性判别分析(Probabilistic linear discriminant analysis)的表情识别方法SS-PLDA.该方法由两部分构成:第一部分是使用稀疏的方法挑选出人脸与表情相关的区域,构造表情的完备特征集;第二部分是针对构造的表情完备特征集里仍含有一些其他信息,运用概率线性判别分析实现表情特征与干扰信息的分离,学习出一个只含有表情信息的子空间,最后基于该表情子空间进行表情识别分析.通过在CK+和JAFFE这两个数据库上面的实验,证实了基于稀疏特征挑选的方法可以得到识别性能的改善,且先使用特征挑选再对所挑选结果应用概率线性判别分析可以达到更好的提升效果.

关键词: 人脸表情识别, 稀疏, 特征挑选, 子空间学习

Abstract: A facial expression recognition method named SS-PLDA is proposed based on Sparse Feature Selection and Probabilistic Linear Discriminant Analysis.The SS-PLDA method contains two steps:1) pick out the most discriminative regions for facial expressions and use these regions to construct a complete facial features set;2) apply Probabilistic Linear Discriminant Analysis Method to separate the useful expression signals from other disturbance information.Therefore,a subspace which only contains expression information is learnt and the expression recognition task is implemented in this subspace.Experimental results on Cohn-Kanade(CK+) database and JAFFE database show that the complete facial features set can improve the performance,and the proposed method outperform the state-of-art ones.

Key words: facial expression recognition, sparse, feature selection, subspace learning

中图分类号: