Sep 2026· Proceedings of the 34th ACM International Conference on Multimedia· 0 citations· 68 references
Computer Science
TL;DR
Inspired by the forensic workflow of human judicial experts, Agentic Tool-Augmented Reasoning (ATAR) is proposed, a framework integrating 22 specialized forensic tools across seven complementary domains to autonomously detect, localize, and explain image forgeries through multi-turn reasoning.
Abstract
Conventional image forgery detection methods produce only binary scores or pixel-level masks without interpretable evidence, while recent multimodal large language model (MLLM)-based approaches generate textual explanations that are merely post-hoc verbalizations of pre-determined classification results rather than products of genuine reasoning. Inspired by the forensic workflow of human judicial experts---''experimental analysis -- logical reasoning -- scientific evidence''---we propose an Agentic Tool-Augmented Reasoning (ATAR) framework for explainable image forgery detection, integrating 22 specialized forensic tools spanning seven complementary domains to autonomously detect, localize, and explain image forgeries through multi-turn reasoning. Concretely, we propose a Dual-Stream Forensic Reasoning paradigm to emulate the experimental analysis process of forensic experts: (1) a high-level semantic anomaly path, which magnifies suspicious regions for fine-grained inspection; and (2) a low-level forgery artifact path, which invokes forensic tools to extract objective artifactual evidence. To this end, we further propose a Forensics Curriculum Learning training strategy. First, during the General Experience SFT stage, an automated teacher--student mentoring pipeline is designed to synthesize multi-turn tool-usage reasoning trajectories. Subsequently, during the Forensic Scene RL stage, a Tool Prior Curriculum is introduced to guide early tool exploration and progressively transfer 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.
FACT (Forensic Agent with Compiled Tool-use Trajectories), which learns an image-conditioned tool-use policy for forensic analysis, achieves the best performance among all compared methods, including on recent unseen generators, deepfakes, and manipulated images.
Jiao-Yang Chen, Bin Hu, Jingyu Hu et al.· 0 citations
An evidence-guided detector-localizer-reasoner system, where an image-level detector provides a global authenticity prior, a dedicated localizer extracts tampered regions as spatial grounding evidence, and an MLLM-based reasoner generates structured forensic reports grounded in this expert forensic evidence is proposed...
Pei-Feng Liu, Bin Li, Qing-Song Zhang et al.· Proceedings of the 34th ACM...· 0 citations
A decomposed chain-of-thought (CoT) pipeline that combines a document tampering detector (DTD) with two Qwen3-VL-32B vision-language models, each LoRA-adapted to a distinct sub-task is presented.
Kirill Koltsov, Aleksandr Gushchin, D. Vatolin et al.· Proceedings of the 34th ACM...· 0 citations
This work introduces ManiVid, a unified forensic analysis task covering forgery detection, artifact grounding, and anomaly explanation for manipulated videos, and proposes ManiVidLens, a unified framework for explainable video forgery analysis.
Heng-Rui Kang, Zhong-Hao Yan, Yuxuan Yang et al.· arXiv.org· 0 citations
Reliable face forgery detection is critical to the security of online identity verification systems, where missed attacks compromise security and excessive false positives disrupt legitimate users. Specialized forensic detectors achieve strong detection performance but provide limited interpretability, while multimodal...
Hang Zhou, Yi-Ming Tang, Kun Yu et al.· arXiv.org· 0 citations
XAI-VFF is presented, a unified Explainable AI framework that combines Gradient-weighted Class Activation Mapping (Grad-CAM), SHapley Additive exPlanations (SHAP), and Local Interpretable Model-Agnostic Explanations (LIME) for video forgery explanations.
S. P, R. Priya· International journal of com...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.