Face De-Identification Survey Covers Capture to Processing
A new survey on arXiv (2607.25926) provides the first unified overview of face de-identification (De-ID) spanning from physical-domain methods to digital post-processing and sensor-level privacy. The paper categorizes approaches into three domains: physical (wearable accessories that conceal identity before capture), sensor-level (integrating privacy mechanisms during image acquisition), and digital (modifying pixel or appearance features after capture). The work addresses the growing need for privacy in computer vision and responsible AI, aiming to remove identifiable facial features while preserving utility for downstream tasks. The survey bridges sensing systems and vision algorithms, covering both early and contemporary techniques.
Key facts
- Face De-ID aims to remove or conceal personally identifiable facial features.
- The survey covers physical, sensor-level, and digital domains.
- Physical-domain methods use wearable accessories to conceal identity before capture.
- Sensor-level methods integrate privacy mechanisms during image acquisition.
- Digital methods modify pixel-level or appearance-level features post-capture.
- The paper is the first unified overview spanning capture to processing.
- arXiv identifier: 2607.25926.
- The survey addresses data privacy and responsible AI concerns.
Entities
Institutions
- arXiv