Research map / Adversarial robustness and security / Attacks on text and graphs
Morié Attack (MA): A New Potential Risk of Screen Photos
Dantong Niu, Ruohao Guo, Yisen Wang
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 |
|---|---|---|---|
| simulate camera capture and optimize sensor noise to produce a moiré pattern | Inception-V3 · ImageNet validation, targeted attack, ε=4 | attack success rate: 0.970 | Improved |
| simulate camera capture and optimize sensor noise to produce a moiré pattern | Inception-V3 · ImageNet validation, untargeted attack | attack success rate: 1.000 | Improved |
| use moiré-pattern perturbations instead of conventional digital attacks | ImageNet validation, Inception-V3 · untargeted attack, JPEG qf=20 | attack success rate: 0.286 -> 0.938 (BIM); 0.276 -> 0.938 (PGD); 0.491 -> 0.938 (MI-FGSM) | Improved |
| generate perturbations through the camera pipeline rather than adding them after demosaicing (ablation) | Moiré Attack · ImageNet validation, untargeted attack, JPEG compression | attack success rate: 0.857 -> 0.938 (qf=20); 0.902 -> 0.949 (qf=40) | Improved |
| increase sensor-noise budget (ablation) | Moiré Attack · ImageNet validation, targeted attack | attack success rate: 0.876 -> 0.970 -> 0.990 -> 0.995 | Improved |
| increase sensor-noise budget (ablation) | Moiré Attack · ImageNet validation, untargeted attack | attack success rate: 1.000 -> 1.000 -> 1.000 -> 1.000 | No clear effect |
| mean-value normalize the generated examples (ablation) | Moiré Attack · ImageNet validation, untargeted attack, ε=4 | attack success rate: 1.000 -> 0.987 | Hurt |
| apply JPEG compression to generated adversarial examples | Moiré Attack · ImageNet validation, untargeted attack, ε=4 | attack success rate: 1.000 -> 0.938 (qf=20); 1.000 -> 0.949 (qf=40); 1.000 -> 0.956 (qf=60); 1.000 -> 0.9881 (qf=80) | Hurt |
| apply demoiréing preprocessing | Moiré Attack · ImageNet validation, untargeted attack, ε=4 | attack success rate: 1.000 -> 0.989 (MopNet); 0.997 (MDDM); 1.000 (FHde2Net); 0.974 (AMNet) | Mixed |
| transfer the attack from Inception-V3 to other architectures | Moiré Attack · ImageNet validation, ε=4 untargeted attack | attack success rate: 0.967 (ResNet50); 0.927 (ResNet34); 0.959 (DenseNet121); 0.982 (VGG16); 0.973 (VGG19) | Improved |
| increase the monitor display size scale factor (ablation) | Moiré Attack · synthesized moiré pattern | moiré distortion | Improved |
| vary the simulated camera rotation angle (ablation) | Moiré Attack · synthesized moiré pattern | pattern appearance | Mixed |
Closest papers by meaning
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · similarity 0.84)
- PhyCamo: A Robust Physical Camouflage via Contrastive Learning for Multi-View Physical Adversarial Attack (AAAI 2025 · similarity 0.83)
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.82)
- Through the Lens: Benchmarking Deepfake Detectors Against Moiré-Induced Distortions (NeurIPS 2025 · similarity 0.82)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.81)
- PubDef: Defending Against Transfer Attacks From Public Models (ICLR 2024 · similarity 0.81)
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.
