Image Differential Privacy and K-anonymity for Pedestrian Image Data: Impact on Cross-camera Person Re-identification & Demographic Predictions

Abstract

Video cameras are prevalent in large cities but their use outside of public safety remains limited due to legitimate privacy concerns. Nevertheless, the rich information they can capture appears incredibly promising for large-scale smart city applications, as they can function as very powerful and versatile sensors. This ambivalence raises the question of whether such image data can be used in a privacy-responsible manner. Encryption-based solutions assume the end server can be trusted with keeping data safe; data leaks show us this assumption does not necessarily hold true. Traditional image obfuscation methods such as pixelization or blurring on the other hand fail to offer both sufficient privacy and utility. As such, privacy approaches that can provide privacy protection directly on the data itself while retaining practical utility are required. We here extend two such notions, differential privacy and k-anonymity, to image data, and extensively evaluate the resulting privacy-utility trade-off on cross-camera person re-identification and attribute recognition data. Our results show that our proposed approaches can significantly reduce the privacy-sensitivity of image data at source while retaining decent utility for vision-based smart city applications.

Links

code:references.bib

@article{bib:lucas_DiffPrivCam_TCPS2025,

author={Maris, Lucas and Matsuda, Yuki and Yasumoto, Keiichi},

title={Image Differential Privacy and K-anonymity for Pedestrian Image Data: Impact on Cross-camera Person Re-identification \& Demographic Predictions},

journal={Transactions on Cyber-Physical Systems (TCPS)},

volume={2025},

number={},

year={2025},

pages={1--33},

doi={10.1145/3743680}

}

https://scrapbox.io/files/68cb5551c3e8bd5408cc8a52.png

Category

Keywords

Collaborating Organization