Original title: From Detection to Localization: A Unified Forensics Framework for Fully Synthetic and Tampered Images
Annalisa Gallina, Marco Fiorucci, Marco Brigo, Federica Battisti +1 · September 2026
Generative computer models can make realistic fake pictures, making it hard to spot altered media. Most older tools treat this as a simple real-or-fake choice. This setup misses differences between entirely fake images and real ones with altered parts, and it cannot show where edits happened. To fix this, researchers built on an existing tool to create a unified system. It sorts pictures into three groups: real, fully generated, or tampered. It also uses a segmentation branch, which is a tool that inspects an image pixel by pixel to pinpoint edited spots. Compared to selected recent benchmark tools, this method works efficiently, gives better classification accuracy, and scores higher when locating altered areas.
Imagine sorting laundry into whites and colors. Some items are brand new, while others have a stain that's only on one spot. You need to not only separate the items but also point out exactly where each stain is. In this paper, the system does similar work for images: it sorts them into categories and pinpoints the manipulated areas.
Everyone sees images online, and when fakes spread, they can mislead people. This tool helps identify not just if an image is fake, but where it was manipulated, making verification and accountability easier.
The abstract doesn't specify limitations, but it notes that the method extends an existing detector and was evaluated on selected benchmarks, which may not cover all manipulation types or real-world scenarios.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
From Detection to Localization: A Unified Forensics Framework for Fully Synthetic and Tampered Images — Annalisa Gallina, Marco Fiorucci, Marco Brigo et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.