Designing transformations that preserve utility in camera and video data while making re-identification and leakage substantially harder.
This project centers on practical privacy protection for visual data. Instead of relying only on blunt obfuscation or simple redaction, the work explores instance-level transformations that preserve downstream utility while reducing the risk of re-identification.
The core idea is to build privacy-preserving pipelines that are useful enough to be adopted in realistic camera and video workflows, especially in settings where people and vehicles are captured without explicit consent.