CFE-PPAR: Compression-Friendly Encryption for Privacy-Preserving Action Recognition
A new method called CFE-PPAR enables privacy-preserving action recognition in videos while maintaining performance after compression. Previous encryption-based approaches suffered catastrophic drops in recognition accuracy and visual quality when videos were compressed. CFE-PPAR uses secret keys to encrypt videos, which can then be directly processed by a video transformer using parameters transformed by the same keys. Experiments on UCF101 and HMDB51 datasets show CFE-PPAR outperforms prior methods under Motion-JPEG and H.264 compression.
Key facts
- CFE-PPAR is the first compression-friendly encryption method for privacy-preserving action recognition.
- Encrypted videos can be directly recognized by a video transformer using parameters transformed by the same keys.
- Previous encryption methods caused catastrophic decrease in recognition performance and visual quality after compression.
- Experiments were conducted on UCF101 and HMDB51 datasets.
- CFE-PPAR outperformed previous methods under Motion-JPEG and H.264 compression.
Entities
—