Sample Specific Multi-Kernel Metric Learning for Person Re-identification

DOI

10.2991/eee-19.2019.38How to use a DOI?

Keywords

Person re-identification, Sample specific multi-kernel learning, Metric learning

Abstract

The existing metric learning based Person Re-Identification are challenged with large appearance variations across non-overlapping cameras. In this paper, we propose to learn a similarity metric that incorporates robustness against cross views appearance variations. The robustness in metric learning is achieved by learning an adaptive sample specific multi-kernel space for each pair of cross-view images for each person in the training set, referred as SSMK. To this end, we first project features of each person into a multi-kernel feature space, where the subtle features of a person are highlighted to help differentiate this person from others. With the discriminative feature projection of all persons in the learned SSMK space, a robust global metric referred as SSMK-M is learned via Local Fisher Discriminant Analysis (LFDA). During testing, SSMK space is constructed in an incremental way to minimize appearance variations. Experiments show that the proposed approach out- performs most non-deep-learning based approaches on popular benchmarks including VIPeR, GRID and CUHK01.

Copyright

© 2019, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Jun Fang
AU  - Ru-fei Zhang
AU  - Feng Jiang
PY  - 2019/07
DA  - 2019/07
TI  - Sample Specific Multi-Kernel Metric Learning for Person Re-identification
BT  - Proceedings of the 2nd International Conference on Electrical and Electronic Engineering (EEE 2019)
PB  - Atlantis Press
SP  - 226
EP  - 241
SN  - 2352-5401
UR  - https://doi.org/10.2991/eee-19.2019.38
DO  - 10.2991/eee-19.2019.38
ID  - Fang2019/07
ER  -