Russian Scientists Develop AI to Detect Deepfakes in Video Calls
Researchers in St. Petersburg have built a neural network that identifies face-swap deepfakes in real time, aiming to protect users from cybercriminals during online calls.
A team at the St. Petersburg Federal Research Center of the Russian Academy of Sciences has created software capable of spotting deepfakes generated through face-swap technology, where a fake face is overlaid on a real person’s photo or video.
The system uses a neural network to automatically recognize manipulated visuals by analyzing numerous hidden parameters in the image and identifying telltale distortions.
In the project’s first phase, researchers assembled a dataset of more than 30,000 face-swap images to train the neural network. The algorithm was taught to detect a wide range of anomalies—from altered head geometry to distortions in individual facial features, as well as mismatched skin tones and background colors.
According to the developers, this library could be integrated into video call security systems, alerting users in real time if a potential deepfake is detected.
