Research map / Detection and segmentation / CNN features
Exploiting Spatial Structure for Localizing Manipulated Image Regions
Jawadul H. Bappy, Amit K. Roy-Chowdhury, Jason Bunk, Lakshmanan Nataraj, B. S. Manjunath
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 joint patch classification and segmentation training | Conv-LSTM network · NIST patch classification | patch accuracy: 86.76% -> 89.38% | Improved |
| add joint training | Conv-LSTM network · IEEE Forensics Challenge patch classification | patch accuracy: 86.05% -> 87.68% | Improved |
| add joint training | Conv-LSTM network · COVERAGE patch classification | patch accuracy: 79.12% -> 80.06% | Improved |
| add joint patch classification training | Conv-LSTM-Conv network · NIST segmentation | segmentation accuracy: 82.78% -> 84.60% | Improved |
| add joint patch classification training | Conv-LSTM-Conv network · IEEE Forensics Challenge segmentation | segmentation accuracy: 74.32% -> 77.67% | Improved |
| add joint patch classification training | Conv-LSTM-Conv network · COVERAGE segmentation | segmentation accuracy: 78.42% -> 81.14% | Improved |
| add two convolution layers before the LSTM (ablation) | S-LSTM-Conv · NIST segmentation | segmentation accuracy: 78.60% -> 82.78% | Improved |
| add conv5 layer after LSTM (conv3,4 only vs conv3,4,5) (ablation) | S-LSTM network · NIST segmentation | segmentation accuracy: 74.43% -> 78.60% | Improved |
| use LSTM-based network instead of FCN (ablation) | segmentation network · NIST segmentation | segmentation accuracy: 64.34% -> 74.43% | Improved |
| add a convolutional front end before the LSTM for patch classification | P-LSTM · NIST patch classification | patch accuracy: 81.05% -> 86.76% | Improved |
| use 16 feature maps in conv1 instead of 8 (ablation) | J-Conv-LSTM-Conv · NIST segmentation | segmentation accuracy: 80.88% -> 84.60% | Improved |
| use 16 feature maps in conv1 instead of 32 (ablation) | J-Conv-LSTM-Conv · NIST segmentation | segmentation accuracy: 73.17% -> 84.60% | Improved |
Closest papers by meaning
- Training-Free Image Manipulation Localization Using Diffusion Models (AAAI 2025 · similarity 0.88)
- Towards Modern Image Manipulation Localization: A Large-Scale Dataset and Novel Methods (CVPR 2024 · similarity 0.87)
- UnionFormer: Unified-Learning Transformer with Multi-View Representation for Image Manipulation Detection and Localization (CVPR 2024 · similarity 0.86)
- Can We Model the Artifacts Explicitly? Disentangle Artifacts via Pairwise Edit Relations for Image Manipulation Localization (NeurIPS 2026 · similarity 0.86)
- Less Supervision, Better Generalization: Weakly Supervised Fake Region Localization in Diffusion-Edited Images (NeurIPS 2026 · similarity 0.86)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.85)
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.
