电信科学 ›› 2018, Vol. 34 ›› Issue (4): 31-40.doi: 10.11959/j.issn.1000-0801.2018010

• 研究与开发 • 上一篇    下一篇

多稀疏表示分类器决策融合的人脸识别

唐彪,金炜,符冉迪,龚飞   

  1. 宁波大学信息科学与工程学院,浙江 宁波 315211
  • 修回日期:2017-12-14 出版日期:2018-04-01 发布日期:2018-05-02
  • 作者简介:唐彪(1994-),男,宁波大学信息科学与工程学院硕士生,主要研究方向为模式识别、压缩感知和图像处理。|金炜(1969-),男,博士,宁波大学信息科学与工程学院副教授、硕士生导师,主要研究方向为压缩感知、模式识别和数字图像处理等。|符冉迪(1971-),男,宁波大学信息科学与工程学院副教授、硕士生导师,主要研究方向为压缩感知、数字图像处理等。|龚飞(1989-),男,宁波大学信息科学与工程学院硕士生,主要研究方向为机器学习、压缩感知和图像处理。
  • 基金资助:
    国家自然科学基金资助项目(61471212);浙江省自然科学基金资助项目(LY16F010001);宁波市自然科学基金资助项目(2016A610091);宁波市自然科学基金资助项目(2017A610297)

Face recognition using decision fusion of multiple sparse representation-based classifiers

Biao TANG,Wei JIN,Randi FU,Fei GONG   

  1. Faculty of Electrical Engineering and Computer Science,Ningbo University,Ningbo 315211,China
  • Revised:2017-12-14 Online:2018-04-01 Published:2018-05-02
  • Supported by:
    The National Natural Science Foundation of China(61471212);The Natural Science Foundation of Zhejiang Province of China(LY16F010001);The Natural Science Foundation of Ningbo of China(2016A610091);The Natural Science Foundation of Ningbo of China(2017A610297)

摘要:

针对目前人脸识别仍然存在顽健性较差的问题,提出一种多稀疏表示分类器决策融合(FR-MSRC)的人脸识别方法。首先提取3组特征,并训练3个稀疏表示子分类器,然后引入决策融合的思想,根据每个子分类器的分类性能,通过迭代运算过程自适应确定各子分类器的融合权值,最后利用融合权值对多个子分类器的输出结果进行决策,实现不同复杂因素干扰下的人脸识别,分别在Yale B、JAFFE和AR人脸库中进行光照、表情、遮挡以及多类型因素混合干扰实验。实验结果表明,本文提出的方法在复杂的环境中仍保持较高的识别率,顽健性更佳。

关键词: 人脸识别, 稀疏表示分类, 决策融合

Abstract:

A new approach to face recognition combining decision fusion and multiple sparse representation-based classifiers was proposed to improve the robustness of the traditional methods.Different types of facial features were extracted,followed by training multiple sparse representation sub-classifiers,and then decision fusion was used to obtain the recognition result of the system.The significant advantage of the proposed scheme lines in that the final recognition results were not driven by averaging outputs of multiple sub-classifiers,but driven by combining multiple outputs via weighted fusion method.In particular,the fusion weights were adaptively determined by an iterative pro-cedure according to the different classification performance of each sub-classifier.Extensive experiments on Yale B,JAFFE and AR face databases demonstrate that the proposed approach is much more effective than state-of-the-art methods in dealing with lighting changes,expression changes and face occlusion and multi factor mixed interference.

Key words: face recognition, sparse representation-based classifier, decision fusion

中图分类号: 

No Suggested Reading articles found!