计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (13): 147-153.DOI: 10.3778/j.issn.1002-8331.2004-0001

• 模式识别与人工智能 • 上一篇    下一篇

鉴别性非负表示分类及其在人脸识别中的应用

徐然然,吴小俊,尹贺峰   

  1. 江南大学 物联网工程学院,江苏 无锡 214122
  • 出版日期:2021-07-01 发布日期:2021-06-29

Face Recognition via Discriminative Non-negative Representation Based Classification

XU Ranran, WU Xiaojun, YIN Hefeng   

  1. School of IoT Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2021-07-01 Published:2021-06-29

摘要:

非负表示分类器在人脸识别算法中有着突出的表现,但是各类别表示之间存在相关性,这对分类不利。为了解决这一问题,提出了基于鉴别性非负表示的人脸识别算法。在非负约束的基础上,添加正则项,减少类别间相关性;利用交替方向乘子法对变量进行优化;最后将测试样本划分在最小重构误差所对应的类别中。在4个数据集上的实验结果表明,提出的基于鉴别性非负表示的分类识别算法在分类识别精度上超过其他对比算法。

关键词: 稀疏表示, 非负表示, 鉴别信息, 人脸识别

Abstract:

The non-negative representation-based classifier has achieved outstanding performance among face recognition algorithms, however, the correlation of the representations between different categories is detrimental to the classification. It ignores the structural information between different categories. In order to solve this problem, the paper proposes a face recognition algorithm based on Discriminative Non-negative Representation. Firstly, on the basis of non-negative constraints, regularization term is introduced to reduce the correlation of the representations between different categories. Then, the variables are optimized by the Alternating Direction Method of Multipliers(ADMM). Finally, the test sample is classified into the class that leads to the least reconstruction error. The experimental results on four benchmark datasets show that the proposed classification algorithm is superior to compared approaches in terms of recognition accuracy.

Key words: sparse representation, non-negative representation, discriminative information, face recognition