Agentic reasoning improves explainable detection of fake images
Agentic Tool-Augmented Reasoning for Explainable Image Forgery Detection
Computer Vision and Pattern Recognition
Summary
Detecting fake images usually means simply saying if an image is real or fake without explaining why. This paper introduces a new method called ATAR that uses many special forensic tools working together like human experts do to find and explain image fakes step-by-step. It looks for strange areas and specific signs of tampering, then explains its findings. The authors show their method works better than previous approaches and gives clearer explanations about why an image is flagged as fake.
What this means in practice
- •For digital forensics teams: Detect and explain image forgeries with detailed evidence to support legal or investigative decisions using agentic multi-tool reasoning.
- •For social media moderation teams: Identify manipulated images with transparent explanations to help moderate misleading or harmful content on platforms.
Authors
Zhiya Tan, Jing Huang, Changtao Miao, Lin Tan, Xin Zhang, Weiwei Feng, Jianshu Li, Joey Tianyi Zhou
Abstract
Conventional image forgery detection methods produce binary scores or pixel-level masks without interpretable evidence, while recent multimodal large language model (MLLM)-based approaches generate post-hoc explanations of predetermined classification results rather than reasoning from evidence. Inspired by the forensic workflow of human judicial experts, we propose Agentic Tool-Augmented Reasoning (ATAR), a framework integrating 22 specialized forensic tools across seven complementary domains to autonomously detect, localize, and explain image forgeries through multi-turn reasoning. A Dual-Stream Forensic Reasoning paradigm combines a high-level semantic anomaly path, which magnifies suspicious regions for fine-grained inspection, with a low-level forgery artifact path, which invokes forensic tools to extract objective evidence. We further introduce Forensics Curriculum Learning: during General Experience SFT, an automated teacher-student mentoring pipeline synthesizes multi-turn tool-usage reasoning trajectories; during Forensic Scene RL, a Tool Prior Curriculum guides early tool exploration and progressively transfers control to the agent, while a Structured Evidence Reward provides fine-grained process-level supervision. Experiments across IMDL, Deepfake detection, DMDL, and AIGC detection show that ATAR achieves 78.5% average image-level F1 on six zero-shot IMDL benchmarks, surpassing the strongest MLLM baseline by 11.8 percentage points, and remains competitive with specialized detectors on other tasks while producing substantially more faithful and grounded explanations.