Face changing defense method and device for black box face operation model
By training an agent operation model and adding adversarial perturbations, adversarial images are generated to interfere with the black-box face operation model. This solves the problem of poor generalization performance of existing defense strategies and achieves real-time defense against the black-box face operation model with a high success rate.
Patent Information
- Application Number
- CN202511598071.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-04
- Publication Date
- 2026-03-31
- Estimated Expiration
- 2045-11-04
AI Technical Summary
Existing defense strategies have poor generalization performance against face-swapping operations of black-box face manipulation models, and cannot effectively prevent the spread of deepfake images. Furthermore, existing methods are not applicable to attack methods of different face manipulation models.
By training a proxy operation model and adding adversarial perturbations to generate adversarial images, an identity feature extraction network, an identity invalidation network, and an image generation network are used to interfere with the generation process of the black-box face operation model, generating images with different identity content from the source face image. The perturbation is updated using the PGD attack method to adapt to attacks from different models.
It achieves real-time defense against black-box face manipulation models, has good generalization performance, a defense success rate of 94.75%, effectively disrupts identity transfer, and adapts to different face manipulation model attack methods.