Diffusion-Collaborated Text--Visual Multimodal Guidance for Image Manipulation Localization in IoT Supply Chain Systems

AI-powered supply chain systems, a critical component of IoT, leverage automated image verification for deliveries but face vulnerabilities such as image manipulation, which can introduce false data and security risks. With advancements in generative AI, attackers can now manipulate logistics images more convincingly, altering goods’ status, concealing damages, or even erasing items entirely, making detection increasingly difficult. Previous research addressing this issue has typically encountered a decrease in performance when confronted with a variety of manipulated images in real-world situations. To address this issue, in this article, we propose an image manipulation localization framework based on multimodal feature guidance collaborating with advanced latent diffusion models (LDMs). By reconstructing rich intermediate feature representations of images under fake-oriented conditions using diffusion, we combine multimodal fused high-level semantic information to guide the segmentation process, achieving state-of-the-art performance. We further propose a high-quality image manipulation dataset tailored for supply chain scenarios, filling the gap in this field. Extensive experiments further demonstrate the effectiveness and robustness of identifying forged images.