Research map / Adversarial robustness and security / Attacks on text and graphs
A Secure Image Watermarking Framework with Statistical Guarantees via Adversarial Attacks on Secret Key Networks
Feiyu CHEN, Wei Lin, Ziquan Liu, Antoni Chan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use a secret, uniquely trained key network instead of a publicly available pre-trained CNN | zero-bit watermarking · Model-fake attack | false rate (%): 100.00% -> 4.00% | Improved |
| use the secret key network against recursive overwriting with 5 watermarks | zero-bit watermarking · Overlapping watermarks | detection rate of original watermark: 12.00% -> 99.60% vs SSLWM (Ours(S)) | Improved |
| combine length and angle hypothesis tests (ablation) | watermark detector · watermarked images, alpha 0.05 | percent of p-values below 0.05: 95.74% -> 99.48% (length only vs combined) | Improved |
| use L2-norm PGD with gradient value instead of sign and L-infinity (ablation) | watermark embedding attack · robustness to distortions | success detection rate | Improved |
| use single-step FGSM-G instead of iterative PGD (ablation) | watermark embedding attack · robustness to distortions | success detection rate | Hurt |
| add a target length to the adversarial loss (ablation) | watermark embedding · robustness to distortions | success detection rate | Improved |
| use angle-only adversarial loss instead of length plus angle (ablation) | watermark embedding · robustness to distortions | success detection rate | Hurt |
| add data augmentation (rotation and cropping) during watermark embedding (ablation) | watermark embedding · robustness to distortions | success detection rate | Improved |
| use the SKN approach versus baselines under rotation and cropping | zero-bit watermarking · rotation and cropping | Avg DR: cropping 97.90% (DNN0B) vs 90.58%; rotation 86.43% (SSLWM) vs 85.89% | Hurt |
| use the SKN approach instead of SSLWM at equal PSNR 32 | watermarking imperceptibility · MSCOCO test | SSIM: 0.9026 -> 0.9768 | Improved |
Closest papers by meaning
- Safe and Robust Watermark Injection with a Single OoD Image (ICLR 2024 · similarity 0.87)
- Leveraging Optimization for Adaptive Attacks on Image Watermarks (ICLR 2024 · similarity 0.87)
- Watermarks in the Sand: Impossibility of Strong Watermarking for Language Models (ICML 2024 · similarity 0.86)
- ROBIN: Robust and Invisible Watermarks for Diffusion Models with Adversarial Optimization (NeurIPS 2024 · similarity 0.86)
- Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models (CVPR 2025 · similarity 0.86)
- Revisiting the Information Capacity of Neural Network Watermarks: Upper Bound Estimation and Beyond (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
