Media manipulation tools and generation techniques have rapidly proliferated the digital world. For example, the advent of GANs has led to more realistic and seamless manipulations that can fool even experts. However, these approaches rely on large amounts of data and often miss semantic details when attempting complex manipulations and generation. The US Department of Defense (DoD) would like to be able to extract knowledge from media assets and use semantic information to detect manipulations. The goal of DISCOVER is to develop a set of forensics tools to determine the integrity, semantic consistency, source, and intent of media assets such as images, videos, and documents.

Sponsored by DARPA’s MediFor program, DiSPARITY develops deep learning algorithms for identifying digital, physical, and semantic image and video manipulations. Image manipulations include image splicing, copy-move, and image repurposing. DiSPARITY also develops robust, subject-agnostic machine learning methods and systems for deepfake detection.