De-identification of facial videos while preserving remote physiological utility
Savic, Marko; Zhao, Guoying (2023-11-24)
BMVA Press
24.11.2023
Savic, M. & Zhao, G. (2023). De-identification of facial videos while preserving remote physiological utility. 34th British Machine Vision Conference 2023, BMVC 2023, Aberdeen, UK, November 20-24, 2023. Retrieved from https://proceedings.bmvc2023.org/230/
https://rightsstatements.org/vocab/InC/1.0/
© 2023. The copyright of this document resides with its authors. It may be distributed unchanged freely in print or electronic forms.
https://rightsstatements.org/vocab/InC/1.0/
© 2023. The copyright of this document resides with its authors. It may be distributed unchanged freely in print or electronic forms.
https://rightsstatements.org/vocab/InC/1.0/
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:oulu-202312083561
https://urn.fi/URN:NBN:fi:oulu-202312083561
Tiivistelmä
Abstract
Remote photoplethysmography has great potential in future telemedicine and affective computing, but contains sensitive biometric data, making privacy protection essential for future applications. Privacy related research concerning rPPG is severely limited and does not handle utility-preserving de-identification. In this paper we propose the first learning based method for facial video de-identification that preserves the rPPG signal and visual appearance, thus keeping the utility of the data for remote rPPG measure while protecting users' privacy. Our proposed semi-adversarial framework processes an input video by adding unobtrusive perturbations that remove biometric privacy while keeping the rPPG signal quality high. The framework is trained via learning-based constraints that leverage pre-trained biometric recognition networks and rPPG predictors. Furthermore, we propose a novel loss term that improves biometric de-identification by lowering downstream recognition confidence. We systematically evaluate our proposed method on two public datasets and with varied face identification and rPPG extraction methods, and provide a novel benchmark for future research in this direction. Our code is available at: https://github.com/marukosan93/De-id_rPPG
Remote photoplethysmography has great potential in future telemedicine and affective computing, but contains sensitive biometric data, making privacy protection essential for future applications. Privacy related research concerning rPPG is severely limited and does not handle utility-preserving de-identification. In this paper we propose the first learning based method for facial video de-identification that preserves the rPPG signal and visual appearance, thus keeping the utility of the data for remote rPPG measure while protecting users' privacy. Our proposed semi-adversarial framework processes an input video by adding unobtrusive perturbations that remove biometric privacy while keeping the rPPG signal quality high. The framework is trained via learning-based constraints that leverage pre-trained biometric recognition networks and rPPG predictors. Furthermore, we propose a novel loss term that improves biometric de-identification by lowering downstream recognition confidence. We systematically evaluate our proposed method on two public datasets and with varied face identification and rPPG extraction methods, and provide a novel benchmark for future research in this direction. Our code is available at: https://github.com/marukosan93/De-id_rPPG
Kokoelmat
- Avoin saatavuus [34357]