VisionLabs: Technology for fake video recognition

Main article: DeepFake

2023: Developing technology to recognise fake video

The company VisionLabs on June 30, 2023 announced the development of recognition technology deepfakes for detection. With fake content the help neural networks of, the detector finds on video a replacement for a person's face or facial expressions, and also determines the generated faces. The solution will help prevent the spread of false in and. information social networks messengers

Deepfakes are created using neural networks that replace a person's face in the image. At the same time, in the created video there are signs of editing - for example, the difference in lighting and light between the superimposed face and the environment or artifacts invisible to the human eye. It is they who can be discovered by VisionLabs technology. The detector works one frame at a time and detects deepfakes for all faces in the image, if there are more than one.

The accuracy of the VisionLabs detector has been tested against various data. Depending on the quality of images and the variety of deepfakes, it reaches from 92% to 100%, which is comparable to the best world algorithms for recognizing deepfakes. When checking on datacet, which includes the most common types of fakes, the technology shows 96.2% accuracy.

The developed technology will find its application in cases where it is necessary to identify fake content - for example, uploading videos from users on social networks, publishing in media or confirming the identity of speakers at online events. Also, a deepfake detector can be used in digital services - remote service with face biometrics authorization, remote certification or digital justice.

Together with VisionLabs' Liveness technology, the deepfake detector will provide comprehensive protection of biometric systems from spoofing. Liveness protects against physical presentation attacks - printed photos, replay attacks and various masks, including 3D, and deepfake detector - from virtual image spoofing.

Recently, a large number of free or inexpensive services for generating content have appeared, which makes deepfake protection technologies more and more relevant. We see great prospects in their development, in the next couple of years, deepfake detection systems will appear in the industrial operation of companies from various fields - from media holdings to financial organizations. The development of deepfake recognition algorithms and their integration into biometric systems is an investment in the provision of competitive and secure services, "said Dmitry Markov, CEO of VisionLabs.