Secure Watermarking and Traceability in Diffusion Models

NTUST Introduction to Information Security – Final Term Report (Spring 2025)

This repository contains the final course report for Introduction to Information Security at
National Taiwan University of Science and Technology (NTUST).
The report surveys recent techniques for securing diffusion-based generative models using watermarking and traceability strategies.

Course: NTUST Introduction to Information Security

---

Report

---

Topics Covered

• Diffusion models overview
• Latent-space watermark injection
• Pixel-space and steganographic methods
• Model fingerprinting and attribution techniques
• Legal & ethical challenges
• Benchmark criteria (robustness, fidelity, stealth)

---

Taxonomy Overview

Model-level Watermarking
├── Latent-space Watermarking
│   ├── CLUE-MARK (arXiv 2024)
│   ├── Robin (NeurIPS 2024)
│   └── LaWa (arXiv 2024)
└── Adversarial & Semantic Conditioning
    ├── InvisMark (WACV 2025)
    └── PT-Mark (arXiv 2025)

Output-level Traceability
├── Image/Video Watermarking
│   ├── StegaStamp (CVPR 2020)
│   ├── CoSDA (AAAI 2025)
│   └── Tree-Ring Watermark (arXiv 2023)
└── Decoder Inversion & Robustness Defenses
    ├── Gradient-Free Decoder Inversion (NeurIPS 2024)
    └── Watermark Removal vs. Defense Study (arXiv 2024)

Provenance Verification
├── Fingerprinting & Residual Signatures
│   ├── Diffusion Signature Analysis (ICCV 2023)
│   ├── VIDiff (CVPR 2024)
│   └── Stable Signature (arXiv 2023)
├── Identity Embedding
│   └── TraceMark-LDM (arXiv 2025)
└── Multimodal & Benchmark Extensions
    ├── ProMark (CVPR 2024)
    └── SAT-LDM (arXiv 2024)

Ownership & Identity Protection
├── PCDiff: Prompt-Controlled Ownership Binding (arXiv 2025)
├── WaDiff: Watermark-Conditioned Diffusion (arXiv 2024)
└── Aqualora: LoRA-based Model Authentication (arXiv 2024)

---

Surveyed Papers

1. Model-level Watermarking

---

2. Output-level Traceability

---

3. Provenance Verification

---

4. Ownership & Identity Protection