ML Security PapersREVEAL -- Reasoning and Evaluation of Visual Evidence through Aligned Language
Ipsita Praharaj 1,2, Yukta Butala 1, Badrikanath Praharaj 2, Yash Butala 1
Published on arXiv
2508.12543
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
REVEAL's two-pronged VLM-driven approach (holistic + region-wise) generalizes across Photoshop, DeepFake, and AIGC editing domains, outperforming competitive supervised baselines while also providing human-interpretable reasoning.
REVEAL
Novel technique introduced
The rapid advancement of generative models has intensified the challenge of detecting and interpreting visual forgeries, necessitating robust frameworks for image forgery detection while providing reasoning as well as localization. While existing works approach this problem using supervised training for specific manipulation or anomaly detection in the embedding space, generalization across domains remains a challenge. We frame this problem of forgery detection as a prompt-driven visual reasoning task, leveraging the semantic alignment capabilities of large vision-language models. We propose a framework, `REVEAL` (Reasoning and Evaluation of Visual Evidence through Aligned Language), that incorporates generalized guidelines. We propose two tangential approaches - (1) Holistic Scene-level Evaluation that relies on the physics, semantics, perspective, and realism of the image as a whole and (2) Region-wise anomaly detection that splits the image into multiple regions and analyzes each of them. We conduct experiments over datasets from different domains (Photoshop, DeepFake and AIGC editing). We compare the Vision Language Models against competitive baselines and analyze the reasoning provided by them.
Key Contributions
- REVEAL framework that frames image forgery detection as a prompt-driven visual reasoning task using semantic alignment capabilities of VLMs
- Holistic Scene-level Evaluation approach that assesses physics, semantics, perspective, and realism of the image as a whole
- Region-wise anomaly detection that partitions images into sub-regions for localized forgery analysis
🛡️ Threat Analysis
Primary contribution is detecting manipulated/AI-generated visual content (deepfakes, AIGC editing, Photoshop forgeries) — a canonical AI-generated content detection task under output integrity. The paper proposes a novel detection methodology (REVEAL) rather than merely applying existing detectors to a specific domain.