Sber received two patents from the Federal Service for Intellectual Property for technologies created as part of a study to detect deepfakes, the company’s press center reported.
1️⃣ The technology is based on a number of ensembles of neural network models of the EfficientNet class (patent No. 2768797)
2️⃣And the method of amplification and analysis by means of AI of microchanges in the colors of objects on frames (patent No. 2774624)
The efficiency indicators of the system, when measured on independent test samples [not specified], amounted to 98%, which is significantly higher than the indicators of published analogues.
The developed technologies will help the company to solve the following tasks:
– Protection against cyber attacks bypassing Face Recognition and Liveness Detection systems;
– Detection of information attacks in the early stages in order to combat fake news;
– Ensuring the protection of negotiations over video conferencing.
“An additional effect of the application of the invention may be its use in biometric control systems, for example, when receiving services (for example, banking services) or access (access control system, turnstile with a biometric sensor). If a synthetically modified face image is detected, access is blocked or the requested action on the part of the user. In this case, user authentication data may be additionally requested, selected from the group: login, code, password, two-factor authentication, or combinations thereof” – patent No. 2768797
The patent containing the method of amplification and analysis by means of AI of microchanges in the colors of objects on frames was recognized by the Federal Service for Intellectual Property as a patent of the week.