Alphabet's Jigsaw subsidiary has unveiled a new tool that aims to help media outlets detect deepfakes and altered images more easily. Assembler, this is the name of the new instrument, makes use of elements generically called "detectors" to analyze an image, a deepfake video, determine its authenticity and, if not, notify in which parts the image or video has been altered .
Jigsaw CEO Jared Cohen explains the mechanism of Assembler in a post on Medium: "Assembler brings together multiple image manipulation detectors from different universities into one tool, each designed to identify specific types of manipulation. Individually these detectors for example, they can identify copy-paste areas or variations in the brightness of the image. Together they are able to carry out a comprehensive check whether an image has been manipulated in any way ". To contribute to the realization of Assembler, prestigious universities around the world participated, including lFederico II University of Naples with Luisa Verdoliva, Associate Professor for the Department of Industrial Engineering.
Assembler, which apparently will not be a tool intended for the wider public, already used in the experimental phase at some information and fact-checking agencies such as Animal Politico, Rappler, Agence France-Presse, Code for Africa and Les Dcodeurs du Monde.
In the past we had already seen something similar with one of Adobe's "Sneaks", the About Face project. Presented last November at Adobe MAX, About Face a tool capable of detect whether a face has been manipulated with digital imaging techniques, indicating how likely the image may have been altered. About Face capable of recognizing which parts of the image have been manipulated, showing a sort of "heatmap" of the areas subject to alteration. Assembler seems to promise the same thing, but more broadly and not focused solely on people's faces.
