Research map / Image generation / Image restoration
NOWA: Null-space Optical Watermark for Invisible Capture Fingerprinting and Tamper Localization
Edwin Vargas, Jhon Lopez, Henry Arguello, Ashok Veeraraghavan
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 |
|---|---|---|---|
| feed null-space projection to the detector instead of the protected image (ablation) | NOWA detector · Tamper localization (COCO) | F1: 0.75 -> near-perfect (F1~0.75, AUC~0.82, IoU~0.69 with image input) | Improved |
| add a learned phase mask (vs no mask) (ablation) | NOWA · Tamper localization | F1: F1=0.89, IoU=0.81 without mask | Improved |
| embed optical null-space watermark vs EditGuard | NOWA · Tamper localization, Stable Diffusion edits | F1: 0.966 -> 0.993 | Improved |
| embed optical null-space watermark vs EditGuard | NOWA · Tamper localization, SDXL edits | F1: 0.965 -> 0.929 | Hurt |
| embed optical null-space watermark vs EditGuard | NOWA · Tamper localization, Lama edits | IoU: 0.934 -> 0.933 | No clear effect |
| NOWA vs EditGuard under JPEG Q=70 | NOWA · Tamper localization under JPEG | F1: 0.552 -> 0.885 | Improved |
| NOWA vs EditGuard under Gaussian noise sigma=5 | NOWA · Tamper localization under noise | F1: 0.871 -> 0.982 | Improved |
| NOWA vs purely digital HiFi-Net under degradation | NOWA · Tamper localization under degradation | F1: 0.156 -> 0.982 (sigma=5) | Improved |
| NOWA vs purely digital detectors (MVSS-Net, OSN, PSCC-Net) | NOWA · Tamper localization under AIGC edits | F1: F1<0.2 for digital detectors -> 0.993 | Improved |
| attacker imitates camera outputs with generative model | NOWA verification · Adversarial robustness, Attack 1 | F1: detection F1 0.901 | Improved |
| attacker imitates protected images with generative model | NOWA verification · Adversarial robustness, Attack 2 | F1: detection F1 0.913 | Improved |
| attacker estimates PSF by blind deconvolution | NOWA verification · Adversarial robustness, Attack 3 | F1: detection F1 0.946 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EditGuard — EditGuard: Versatile Image Watermarking for Tamper Localization and Copyright Protection (CVPR 2024) · better on 10 of 12 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OmniGuard: Hybrid Manipulation Localization via Augmented Versatile Deep Image Watermarking (CVPR 2025 · similarity 0.83)
- Watermarking One for All: A Robust Watermarking Scheme Against Partial Image Theft (CVPR 2025 · similarity 0.82)
- StableGuard: Towards Unified Copyright Protection and Tamper Localization in Latent Diffusion Models (NeurIPS 2025 · similarity 0.82)
- EditGuard: Versatile Image Watermarking for Tamper Localization and Copyright Protection (CVPR 2024 · similarity 0.82)
- FreqMark: Invisible Image Watermarking via Frequency Based Optimization in Latent Space (NeurIPS 2024 · similarity 0.80)
- Safe-Sora: Safe Text-to-Video Generation via Graphical Watermarking (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.
