SaM-fake is the successor project to MobileTrust and is dedicated to addressing the challenge of combating misinformation and manipulation in audiovisual content spread through social media. Our approach integrates invisible signatures directly into the medium during recording, which remain intact even after compression, encoding, rotation, or changes in image size. This technology aims to enable reliable detection of manipulations, including deep fakes.
Through a special mathematical transformation — using, among other things, FaR hashes and Singular Value Decomposition (SVD) — media content is transformed into a non-directly perceivable domain during recording. The information is embedded as FaR hashes into the SVD matrices and then re-transformed back into the visible medium. This distribution of the signature across the entire image or audio, combined with error-correcting encoding, ensures high robustness against legitimate post-processing, such as compression or rotation.
The goal of the project is to develop a quality-assured software product that enables court-admissible image and sound recordings and supports the signing of AI-generated content. Additionally, a community version of the software will be integrated into messenger apps to directly distribute verified content on social networks, thus making the spread of fake news and deep fakes more difficult.
Potential user groups include public media, reporters, regulatory authorities, and industries that rely on tamper-proof recordings, such as medical technology or aircraft manufacturing. By combining innovative mathematical methods with a user-friendly software solution, the project offers great potential to sustainably ensure the authenticity of audiovisual content in digital media.
We are supported in this project by our industry partner Chainstep GmbH.
