A face image attribute protection method based on adaptive confrontation
By introducing a natural loss function and an adaptive PGD algorithm into the total loss function, and combining momentum mechanism and SMAC algorithm to optimize the attack step size, the generality and adaptability problems of existing face image attribute protection methods are solved, achieving effective protection against various face tampering attacks and maintaining the visual naturalness of images.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG POLYTECHNIC NORMAL UNIV
- Filing Date
- 2026-02-05
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for protecting facial image attributes suffer from poor versatility, lack of adaptive adjustment capabilities, and insufficient visual naturalness. They are difficult to adapt to various facial tampering attack models, and fixed perturbation parameters lead to unstable protection effects and an imbalance in image usability.
An adaptive adversarial face image attribute protection method is adopted. By introducing a natural loss function into the total loss function, the perturbation is dynamically adjusted using an adaptive PGD algorithm. The attack step size is optimized by combining momentum mechanism and SMAC algorithm, and image-level and model-level perturbation fusion is performed to achieve adaptive adjustment of adversarial perturbation and preservation of visual naturalness.
It achieves adaptive protection against various face tampering attack models, maintains the visual naturalness and usability of images, and improves the generalization ability and stability of the defense.
Smart Images
Figure CN122415306A_ABST