AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Creating Blank Canvas Against AI-enabled Image Forgery

Qi Song, Ziyuan Luo, Renjie Wan

AAAI 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Neural network watermarking · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
protect images with blank-canvas adversarial perturbation before tampering (ablation)SAM tamper localization · Ablation (tamper localization)F1: 0.352 -> 0.964Improved
add frequency-aware optimization (L_stealth plus adaptive spectral step) to MSE attack (ablation)blank canvas perturbation · Ablation (tamper localization)F1: 0.934 -> 0.964Improved
add frequency-aware optimization (ablation)blank canvas perturbation · Ablation (tamper localization)IoU: 0.928 -> 0.955Improved
add random degradation to protected images before tampering (ablation)full method · Ablation (tamper localization)F1: 0.964 -> 0.945Hurt
use proactive blank-canvas protection instead of passive detectorsAIGC-edited images (AGE-Set) · AGE-Set SD InpaintF1: 0.694 -> 0.972Improved
blank-canvas SAM vs EditGuard on classical benchmarkstamper localization · CASIA1+F1: 0.67 -> 0.67No clear effect
blank-canvas SAM vs EditGuard on AIGC editstamper localization · AGE-Set RePaintF1: 0.967 -> 0.961Hurt
blank-canvas SAM vs EditGuard on Korustamper localization · KorusF1: 0.28 -> 0.31Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy