RECOGNITION OF FAKE (SPURIOUS) VIDEO CONTENT SYNTHESIZED USING DEEPFAKE TECHNOLOGY OF THE GENERATIVE ADVERSARIAL NETWORK (GAN) ALGORITHM

Dzhurov Alexander Andreevich, Cherckesova Larisa Vladimirovna, Revyakina Elena Aleksandrovna

In the modern world, one of the main and actual problems is fake (false) content: news, videos, photos, etc. At the early stage of the development of DeepFake imitation technology, it was used mainly by amateur users to synthesize entertaining multimedia content by comparing people's facial expressions and phrases said, as a rule, by recognizable personalities, to create fake news that looks authentic. But the political situation has changed, and DeepFake technology has been used not only to compromise undesirable persons, but also for disinformation and political agitation, as an integral part of the information war. The purpose of study: software implementation of the video content recognition algorithm synthesized using the DeepFake technology of the Generative Adversarial Networks (GAN) algorithm with acceptable correctness and accuracy. The paper proposes the software implementation that analyzes video content and makes decision about its authenticity. The main architectures of the GAN algorithm are presented; the results and consequences of using DeepFake technology are considered. The analysis of the features of the Xception and ResNeXt models trained using neural networks is carried out. Methods: for the operation of the system, the appropriate neural networks were selected based on the results of their productivity. The software implementation uses ResNeXt and XceptionNet models, as well as a pre-trained human face recognition model BlazeFace, used for face recognition on extracted images. Results: the Deep_Fake_Recognizer–23 software tool has been created that recognizes fake video content synthesized using DeepFake technology using the GAN algorithm with acceptable correctness and accuracy.

Keywords:Information war, false content, DeepFake, Generative Adversarial Networks (GAN) algorithm; neural networks, discriminator