Research map / Image generation / Autoregressive and latent generation
AuthSig: Safeguarding Scanned Signatures Against Unauthorized Reuse in Paperless Workflows
Ruiqiang Zhang, Zehua Ma, Guanjie Wang, Chang Liu, Hengyi Wang, Weiming Zhang
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 both content and style features instead of style only (ablation) | AuthSig · GPDS signatures | FID: 370.5272 -> 40.8138 | Improved |
| use both content and style features instead of content only (ablation) | AuthSig · GPDS signatures | FID: 223.3530 -> 40.8138 | Improved |
| add both features: HWD (ablation) | AuthSig · GPDS signatures | HWD: 1.7905 -> 0.2047 (vs style only) | Improved |
| fine-tune the VAE on signature data (ablation) | AuthSig · GPDS 1-150 | rFID: 5.76 -> 1.63 | Improved |
| fine-tune the VAE on signature data (ablation) | AuthSig · GPDS 1-150 | rHWD: 0.1446 -> 0.0344 | Improved |
| fine-tune the VAE on signature data (ablation) | AuthSig · IAM | rFID: 5.02 -> 0.80 | Improved |
| embed watermark in disentangled style features vs prior watermarking (CIN) under background transparency | AuthSig · GPDS signatures | extraction accuracy: 0.7195 -> 0.9921 | Improved |
| high brightness factor of 2.0 | AuthSig · brightness distortion | extraction accuracy: 93.7% at brightness 2.0 | Hurt |
| embed watermark via style modulation in generative model vs pixel-domain methods under print-scan | AuthSig · Print-scan | extraction accuracy: above 97% for most device pairs | Improved |
Closest papers by meaning
- PP-Mark: Provable and Publicly Verifiable Watermarking for Generative AI (NeurIPS 2026 · similarity 0.83)
- ImageSentinel: Protecting Visual Datasets from Unauthorized Retrieval-Augmented Image Generation (NeurIPS 2025 · similarity 0.82)
- Beyond Bit Matching: Orthogonal Watermarks for Collusion-Resistant Image Fingerprinting (NeurIPS 2026 · similarity 0.82)
- Disentangled Style Domain for Implicit $z$-Watermark Towards Copyright Protection (NeurIPS 2024 · similarity 0.81)
- Can Simple Averaging Defeat Modern Watermarks? (NeurIPS 2024 · similarity 0.80)
- Transferable Black-Box One-Shot Forging of Watermarks via Image Preference Models (NeurIPS 2025 · similarity 0.80)
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.
