基于字典学习和Fisher判别稀疏表示的行人重识别方法

Pedestrian Re-Identification on the Basis of Dictionary Learning and Fisher Discrimination Sparse Representation

针对目前的字典学习方法对不同摄像机视角行人特征的联系考虑不足的问题,提出了一种新的基于字典学习和Fisher判别稀疏表示的行人重识别方法.该方法考虑不同场景中同一行人的特征应该具有相似的稀疏表示,提出行人重识别离散度函数的概念,加入约束稀疏表示的正则化项,最大化不同行人稀疏表示的类间离散度,同时最小化同一行人稀疏表示的类内离散度,通过学习到的字典得到具较强区分识别能力的稀疏表示.在公开数据集VIPeR、PRID 450s和CAVIAR4REID上的实验表明,文中方法的识别率高于目前基于字典学习的行人重识别方法.

In order to overcome the inadequate consideration of the existing dictionary learning taken into the con-nection of pedestrian features of different camera views, a new pedestrian re-identification method is proposed on the basis of dictionary learning and Fisher discrimination sparse representation. By considering the similar sparse representation of features of the same pedestrian in different scenes, the concept of pedestrian re-identification scat-ter function is put forward through adding a regularization term that constrains the sparse representation. The regu-larization term aims at maximizing the between-class scatter of the sparse representation of different pedestrians, and minimizing the within-class scatter of the sparse representation of the same pedestrian. Thu s , sparse representation with strong discrimination ability can be obtained via dictionary learning. Experimental results on VIPeR, PRID 450s and CAVIAR4REID datasets indicate that the recognition rate of the proposed method is higher than that of other dictionary learning-based pedestrian re-identification methods.