中文摘要：

该文提出了基于自适应度量学习（AML）的行人再识别方法。与正常处理所有负样本的常规度量学习方法不同的是,AML将负样本自适应地分为三组,并对它们给予不同的关注。通过加强负样本的影响,AML可以更好地挖掘正样本和负样本之间的辨别信息,从而生成更有效的度量。此外,我们还提出了探针特定重新排名（PSR）算法来改进由度量学习得到的初始排名列表。对于每个探针,PSR构建相应的超图以捕获探针和其排名前100的图库图像之间的邻域关系。然后基于它们在超图中的邻域亲和力来重新排列这些图像。其中对公共数据集VIPeR数据集的实验证明了AML和PSR的良好的鲁棒性和优越性。

英文摘要：

This paper presents a pedestrian re-recognition method based on adaptive metric learning（AML）. Unlike conventionalmeasurement methods that normally handle all negative samples, AML adaptively divides negative samples into three groups andgives them different concerns. By enhancing the impact of negative samples, AML can better exploit the identification informationbetween positive and negative samples, resulting in more efficient measures. In addition, we propose a probe-specific re-rank（PSR） algorithm to improve the initial ranking list obtained from metric learning. For each probe, the PSR constructs a correspond-ing hypergraph to capture the neighborhood relationship between the probe and its top 100 library image. These images are then re-arranged based on their neighborhood affinity in the hypergraph. The experiment of the VIPeR dataset of the public data set provesthe good robustness and superiority of AML and PSR.