一种基于可逆神经网络的人脸身份变换方法
By using a face identity transformation method based on reversible neural networks, identity features are extracted and random keys are generated. Anonymous identity features are generated using affine coupling blocks, which solves the balance problem between anonymization and original identity recovery, and achieves diversified anonymization and secure recovery effects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV OF POSTS & TELECOMM
- Filing Date
- 2023-12-21
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to protect facial privacy while ensuring the security and usability of intelligent systems, particularly in striking a balance between anonymization and restoration of the original identity.
A face identity transformation method based on reversible neural networks is adopted. By extracting identity features, differentiating them into two identity vectors, generating a random key, and using affine coupling blocks to generate anonymous identity features, the network is trained through multi-task contrastive learning to ensure the reversibility of anonymization and restoration.
It enables diverse generation and secure restoration of anonymized images. The generated anonymous identities are difficult for attackers to identify, and the original identities can be securely restored when necessary, ensuring user security.
Smart Images
Figure CN117744146B_ABST