Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (8): 158-165.DOI: 10.3778/j.issn.1002-8331.1904-0235

Previous Articles     Next Articles

Normality Resampling of Improved Metric Learning Method for Person Re-Identification

SONG Lili, LI Bin, ZHAO Junya, LIU Guofeng   

  1. 1.The Engineering & Technical College of Chengdu University of Technology, Leshan, Sichuan 614000, China
    2.School of Mechanical Engineering, Wuhan Polytechnic University, Wuhan 430023, China
    3.School of Science, Wuhan University of Technology, Wuhan 430070, China
  • Online:2020-04-15 Published:2020-04-14

正态重采样的改进行人再识别度量学习算法

宋丽丽,李彬,赵俊雅,刘国峰   

  1. 1.成都理工大学 工程技术学院,四川 乐山 614000
    2.武汉轻工大学 机械工程学院,武汉 430023
    3.武汉理工大学 理学院,武汉 430070

Abstract:

For person re-identification task cross camera view, existing metric learning based methods are biased to estimation of model parameters due to the small sample size problem. In this paper, it proposes a normality resampling based cross-view quadratic discriminant analysis method which is a semi-supervised metric learning based method. Moreover, the proposed method is tested on the publicly released datasets VIPeR and CUHK01. The test results show that the proposed method is better than existing cross-view quadratic discriminant analysis method. Especially on the rank-1 identification rate, the proposed method overcomes the state-of-the-art comparative methods MLAPG and NFST method 7.79% and 4.68% respectively. Besides, the proposed method is more robust to the training data size.

Key words: person re-identification, metric learning method, semi-supervised learning, cross-view quadratic discriminant analysis, statistical inference, recognition accuracy

摘要:

跨场景的行人再识别任务,现有度量学习算法由于小样本问题使得对模型参数的估计存在偏差,从而导致识别精度较低。在交叉二次判别分析度量学习算法的基础上,提出了一种基于样本正态性重采样算法,建立了半监督学习度量模型,以增强度量模型的泛化能力。综合泛化后的度量模型和交叉二次判别算法,构建了加权组合的联合模型。选取了公开数据集VIPeR和CUHK01进行测试,测试结果显示该算法相比于原交叉二次判别算法以及相关的行人再识别算法有着明显的优势,尤其在rank-1上的识别精度分别超过了MLAPG算法和NFST算法7.79%和4.68%,且该算法对于训练数据量的变化具有较强的鲁棒性。

关键词: 行人再识别, 度量学习算法, 半监督学习, 交叉二次判别分析, 统计推断, 识别精度