Project Page

HTML Editor - Full Version

Person Re-Identification using Kernel-based 

Metric Learning Methods

Fei Xiong       Mengran Gou       Octavia Camps        Mario Sznaier

Northeastern University, Boston


Re-identi cation of individuals across camera networks with limited or no overlapping elds of view remains challenging in spite of significant research e orts. In this paper, we propose the use, and extensively evaluate the performance, of four alternatives for re-ID classification: regularized Pairwise Constrained Component Analysis, kernel Local Fisher Discriminant Analysis, Marginal Fisher Analysis and a ranking ensemble voting scheme, used in conjunction with different sizes of sets of histogram-based features and linear, X2 and RBF-X2  kernels. Comparisons against the state-of-art show significant improvements in performance measured both in terms of Cumulative Match Characteristic curves (CMC) and Proportion of Uncertainty Removed (PUR) scores on the challenging VIPeRiLIDS, CAVIAR and  3DPeS datasets.