Research map / Detection and segmentation / CNN features
Frequency Mining Empowered by Text Aggregation: A New Perspective on Document Image Tampering Detection
Ziqi Yi, Guitao Xu, Shihang Wu, Peirong Zhang, Lianwen Jin
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 |
|---|---|---|---|
| add Visual-Frequency Integration Module (VFIM) (ablation) | TAFE-Net · DocTamper (avg of Test, FCD, SCD) | IoU: 82.6 -> 84.2 | Improved |
| add Multi-Frequency Feature Extractor (MFFE) (ablation) | TAFE-Net · DocTamper (avg of Test, FCD, SCD) | IoU: 72.9 -> 84.2 | Improved |
| add Direction-aware Frequency Decoupling Enhancement (DFDE) (ablation) | TAFE-Net · DocTamper (avg of Test, FCD, SCD) | IoU: 80.1 -> 84.2 | Improved |
| use DFDE instead of FFDN's WFE (ablation) | TAFE-Net · DocTamper (avg of Test, FCD, SCD) | IoU: 81.9 -> 84.2 | Improved |
| include vertical (LH) and horizontal (HL) subbands in DFDE (ablation) | TAFE-Net · DocTamper (avg of Test, FCD, SCD) | IoU: 82.2 -> 84.2 | Improved |
| use asymmetric convolutions instead of square convolutions in DFDE (ablation) | TAFE-Net · DocTamper (avg of Test, FCD, SCD) | IoU: 83.2 -> 84.2 | Improved |
| use all three frequency inputs (high, low, DCT) versus none (ablation) | TAFE-Net · DocTamper (avg of Test, FCD, SCD) | IoU: 72.9 -> 84.2 | Improved |
| add VFIM, MFFE and DFDE to SegFormer baseline (ablation) | SegFormer · DocTamper (avg of Test, FCD, SCD) | IoU: 68.8 -> 84.2 | Improved |
| swap SegFormer backbone for ConvNeXt V2 (TAFE-Net*) | TAFE-Net · RTM | IoU: 23.1 -> 17.8 | Hurt |
| swap SegFormer backbone for ConvNeXt V2 (TAFE-Net*) | TAFE-Net · DocTamper-Test | IoU: 89.4 -> 90.3 | Improved |
| swap SegFormer backbone for ConvNeXt V2 (TAFE-Net*) | TAFE-Net · DocTamper-SCD | IoU: 74.0 -> 75.9 | Improved |
| TAFE-Net vs FFDN on cross-domain DocTamper-SCD | TAFE-Net · DocTamper-SCD | IoU: 75.2 -> 74.0 | Hurt |
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.
- SegFormer — SegFormer: A Topic Segmentation Model with Controllable Range of Attention (AAAI 2023) · better on 4 of 4 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
- Revisiting Tampered Scene Text Detection in the Era of Generative AI (AAAI 2025 · similarity 0.84)
- TextShield-R1: Reinforced Reasoning for Tampered Text Detection (AAAI 2026 · similarity 0.84)
- Toward Real-world Text Image Forgery Localization: Structured and Interpretable Data Synthesis (NeurIPS 2025 · similarity 0.83)
- Harnessing Frequency Spectrum Insights for Image Copyright Protection Against Diffusion Models (CVPR 2025 · similarity 0.81)
- Towards Better Robustness Against Natural Corruptions in Document Tampering Localization (AAAI 2025 · similarity 0.81)
- Black-Box Detection of Language Model Watermarks (ICLR 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.
