Exploiting color cues to improve person re-identification

• Published in: 7th International Conference on Imaging for Crime Detection and Prevention (ICDP 2016) p. 21
• Main Authors: Chahla, C, Snoussi, H, Abdallah, F, Dornaika, F
• Format: Conference Proceeding
• Language: English
• Published: Stevenage, UK IET 2016; The Institution of Engineering & Technology
• Subjects: Color; Computer vision and image processing techniques; Image recognition; Quaternions; Video signal processing; image classification; low-level color information; feature descriptor; color component; video cameras; object detection; color cue; video surveillance security application; simple metric learning method; nonoverlapping camera; feature extraction; VIPeR dataset; person reidentification; semantic analysis; image colour analysis; video surveillance
• ISBN: 1785614002; 9781785614002
• DOI: 10.1049/ic.2016.0089

Person re-identification is essentially the task of recognizing the same person across different non-overlapping cameras. It has been intensively studied due to its critical role for many security applications of video surveillance. In this paper we exploit the low-level color information in two different aspects, showing the strength provided by colors to increase the accuracy of classification. First, we propose a discriminant embedding for the feature descriptor, which takes into consideration all color components at one time using the quaternions. Second, we propose to assign each individual to a color name in order to increase the discrimination of our descriptor with a semantic analysis. Experiments are carried out on the highly challenging VIPeR dataset. Comparison with some state-of-the-art methods is provided and the proposed method shows better performance even with a simple metric learning method.