Re-Ranking Person Re-Identification with Forward and Reverse Sorting Constraints

摘要

Person re-identification task aims at matching pedestrian images across multiple camera views. Extracting more robust feature of the pedestrian images and finding more discriminative metric learning are the main research directions in person re-identification. The achieved results are provided in the form of a list of ranked matching persons. It often happens that the true match which should be in the first position is not ranked first. In order to correct some false matches and improve the accuracy of person re-identification, this paper proposes a re-ranking method with forward and reverse sorting constraints. The forward sorting constraint makes the image, which is in the front position of one forward sorting list, be backward in the position of other forward sorting lists; The reverse sorting constraint makes two images of the same pedestrian be in the front position of each other's sorting list. Experiments on four public person re-identification datasets, VIPeR, PRID450S, CUHK01 and CUHK03 confirm the simplicity and effectiveness of our method.