Researchers at Cornell University in Ithaca, New York, have proposed a new way to detect video fakes. They have developed a technique that embeds hidden signals in light sources, making it possible to detect even the most sophisticated deepfakes.
Researchers at Cornell University in Ithaca, New York, have proposed a new way to detect video fakes. They have developed a technique that embeds hidden signals in light sources, making it possible to detect even the most sophisticated deepfakes.
As TechRadar reports, the method is called noise-coded illumination. It works on the principle of a watermark, but instead of graphic labels, it uses controlled «noise» in the light. Hidden within these fluctuations is a low-resolution but accurate copy of the scene with timestamps. If the video is altered, the coded light reveals inconsistencies. Such as with deepfakes, editing, or even changing the playback speed.
The technology can be implemented both through software for displays and through a small chip in the lamps. In this case, outsiders are not able to reproduce the signals without a special key. Scientists emphasize that even if a forger learns the algorithm, he would have to synchronize several parallel versions of the video with a hidden light code, which is almost impossible.
The system was tested in a variety of conditions: low light, file compression, camera movement, and both indoors and outdoors. In all cases, the technology remained effective, even when changes were too subtle for the human eye to notice.
According to the developers, the approach could become part of comprehensive cybersecurity systems, for example, to protect surveillance cameras, official records or personal videos that can be the subject of fraudulent manipulation. Unlike traditional checksums, which only record a file change, light codes make it possible to distinguish secure compression from intentional forgery.
The problem of detecting deepfakes has become more acute in recent years due to the availability of powerful editing tools. Similar watermarking technologies are being developed by Google and OpenAI for generative images, but they mostly require control over the source of the file. Cornell offers a solution that can work directly in the camera or light source, making forgeries much more difficult.
We also recently wrote about how 22 Telegram channels were discovered in Poland that spread fake news and propaganda against Ukraine. All of them work for a Polish audience with the aim of hindering support for Ukraine, polarizing public opinion, and creating a split between Polish and Ukrainian societies.
