Computer Engineering and Applications ›› 2017, Vol. 53 ›› Issue (12): 185-189.DOI: 10.3778/j.issn.1002-8331.1601-0064

Previous Articles     Next Articles

Person re-identification based on feature fusion

ZHANG Gengning, WANG Jiabao, ZHANG Yafei, XU Yulong, MIAO Zhuang   

  1. College of Command Information System, PLA University of Science & Technology, Nanjing 210007, China
  • Online:2017-06-15 Published:2017-07-04

基于特征融合的行人重识别方法

张耿宁,王家宝,张亚非,徐玉龙,苗  壮   

  1. 解放军理工大学 指挥信息系统学院,南京 210007

Abstract: Existing methods in person re-identification cannot describe the pedestrian well for significant intra-class variations in illumination and viewpoint. In order to solve this problem, a method based on feature fusion is proposed. Firstly, the image is pre-processed using the Rentiex algorithm. Then the CN feature is combined with the color and texture features. Histograms are extracted through regional and block partition, and then the feature of image is obtained. Finally, person re-identification is performed on four public datasets with different metric learning methods. The experimental results show that the feature combined with CN achieves a stronger representation and increases the accuracy obviously.

Key words: person re-identification, CN feature, feature fusion, histogram

摘要: 针对行人重识别中已有方法难以解决行人图像光照、视角变化大的问题,提出了一种基于特征融合的行人重识别方法。首先利用Retinex变换对图像进行预处理;然后将CN特征与原有的颜色和纹理特征融合,并通过区域和块划分的方式提取直方图获得图像特征;最后采用不同的距离学习方法在4个数据集上进行行人重识别。实验结果表明,融合后的特征对行人图像具有更好的表述能力,实现了重识别精度的较大提升,验证了方法的有效性。

关键词: 行人重识别, CN特征, 特征融合, 直方图