计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (13): 206-211.

• 图形图像处理 • 上一篇    下一篇

基于拓展稀疏表示模型和LC-KSVD的人脸识别

张建明,何双双,吴宏林,熊  兵,李艺敏   

  1. 长沙理工大学 计算机与通信工程学院,长沙 410114
  • 出版日期:2016-07-01 发布日期:2016-07-15

Face recognition based on extend sparse representation and LC-KSVD

ZHANG Jianming, HE Shuangshuang, WU Honglin, XIONG Bing, LI Yimin   

  1. School of Computer and Communication Engineering, Changsha University of Science and Technology, Changsha 410114, China
  • Online:2016-07-01 Published:2016-07-15

摘要: 为了提高人脸的识别率和识别速度及其识别的鲁棒性,提出了基于拓展稀疏表示模型和LC-KSVD(Label Consist K-SVD)的人脸识别算法。针对字典学习中只包含表示能力没有包含类别信息的问题,在原始的稀疏表示模型中添加了残差向量作为系数修正向量,使得拓展稀疏表示模型具有更强的鲁棒性;在字典学习中添加稀疏编码和分类器参数约束项,通过字典学习同时更新稀疏编码和分类器参数,使字典中包含很好的表示能力和判别分类能力。实验结果表明,基于拓展稀疏表示模型和LC-KSVD的人脸识别具有高识别率和低识别速度,并且有很好的鲁棒性。

关键词: 稀疏表示, 字典学习, 人脸识别, LC-KSVD算法

Abstract: To improve the face recognition rate, speed and robustness, this paper proposes a face recognition algorithm based on extended sparse representation model and LC-KSVD(Label Consist K-SVD). For solving the problem that dictionary learning only contains representation ability but no class information, the algorithm adds residual vector as coefficient amending vector into original sparse representation model, making the extended model have stronger robustness. The algorithm also adds sparse coding and classifier parameter constraints into the process of dictionary learning and updates sparse coding and classifier parameters in the process, making the dictionary possess good representation and discrimination ability. The experimental results show that the algorithm has high recognition rate, low recognition speed and good robustness.

Key words: sparse representation, dictionary learning, face representation, LC-KSVD(Label Consist K-SVD)