Dissecting Agentic Forensics: The Role of Triage, Prompting, and Evidence Arbitration in Open-World Fake Image Detection
作者: Xianlong Li, Pietro Bongini, Niccoló Pancino, Marco Blanchini, Benedetta Tondi, Mauro Barni
分类: cs.CV, cs.AI, cs.CR
发布日期: 2026-09-21
备注: Accepted at the 2026 Workshop on AI for Multimedia Forensics and Disinformation Detection (AI4MFDD), ECCV 2026. 34 pages (17 main paper incl. references, 17 appendix), 7 figures, 5 tables
💡 一句话要点
提出基于代理AI的框架以解决开放世界假图像检测问题
🎯 匹配领域: 支柱一：机器人控制 (Robot Control) 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 图像取证 开放世界问题 代理AI 冲突仲裁 推理质量 多模态模型 假图像检测
📋 核心要点
- 现有的图像取证检测器通常专注于单一操控类型，导致在开放世界环境中表现不佳。
- 本文提出了一种无训练的代理框架，结合专业检测器、逐个筛选和冲突感知的证据仲裁，以提高检测性能。
- 实验结果显示，推理质量显著影响检测效果，尤其在分布变化情况下，性能提升明显。
📝 摘要（中文）
图像取证正日益成为一个开放世界问题，现有的取证检测器往往专注于单一的操控类型。代理AI作为一种新兴解决方案，能够评估个别检测器的可靠性、识别超出范围的证据并仲裁冲突报告。本文研究了一种无训练的代理框架，围绕专业检测器、逐个检测器的筛选和冲突感知的证据仲裁展开。结果表明，简单的检测器融合在真实图像上存在严重的误报率，而筛选和提示则能有效提高性能。最重要的是，推理质量是影响性能的主导因素，尤其在分布变化的情况下。开放世界图像取证的主要挑战在于如何校准对专业取证工具的信任及仲裁冲突证据。
🔬 方法详解
问题定义：本文旨在解决开放世界假图像检测中的性能不足问题，现有方法往往无法有效处理多种操控类型和证据冲突。
核心思路：提出一种无训练的代理框架，通过对专业检测器进行逐个筛选和冲突感知的证据仲裁，提升检测的可靠性和准确性。
技术框架：整体架构包括三个主要模块：专业检测器、逐个检测器的筛选和冲突感知的证据仲裁。每个模块协同工作，以提高整体性能。
关键创新：最重要的创新在于引入了推理质量作为性能的主导因素，强调了在分布变化情况下强大的推理能力的重要性。
关键设计：在参数设置上，使用了六种配置和三种多模态大型语言模型作为基础，设计了适应不同数据分布的损失函数和网络结构。通过这些设计，增强了系统的灵活性和适应性。
🖼️ 关键图片
📊 实验亮点
实验结果显示，简单的检测器融合在真实图像上误报率高达严重水平，而通过逐个筛选和提示，性能显著提升。推理质量的提升使得在分布变化情况下，检测性能有了显著改善，特别是在操控召回率几乎饱和的情况下。
🎯 应用场景
该研究的潜在应用领域包括社交媒体内容审核、新闻真实性验证和法律取证等。通过提高假图像检测的准确性和可靠性，能够有效减少虚假信息的传播，增强公众对信息的信任。未来，该框架还可能扩展到其他类型的媒体取证任务中。
📄 摘要（原文）
Image forensics is increasingly an open-world problem: manipulations range from fully synthetic images to localized edits, splicing and swapping, while most forensic detectors remain specialized to a single manipulation family. Agentic AI has recently emerged as a promising solution. In principle, such systems can assess the reliability of individual detectors, identify out-of-scope evidence, and arbitrate conflicting reports. However, it remains unclear which components actually drive performance and whether their benefits persist under distribution shift. To answer these questions, we study a training-free agentic framework built around specialist detectors, per-detector triage, and conflict-aware evidence arbitration. Using six configurations and three multimodal large language model backbones, we dissect the role of triage, prompting, and reasoning quality on both in-distribution and out-of-distribution data. Our results show that naive detector fusion suffers from severe false-positive rates on authentic images. Triage and prompting consistently improve performance by filtering unreliable evidence and exposing detector limitations. However, the dominant factor is represented by reasoning itself: A stronger judge substantially outperforms a weaker one, particularly under distribution shift. Most notably, manipulation recall is nearly saturated across all configurations, indicating that the main challenge of open-world image forensics is not detecting manipulations, but calibrating trust in specialized forensic tools and arbitrating conflicting evidence.