Методологія безпеки нейромережевих інформаційних технологій виявлення deepfake-модифікацій біометричного зображення

Abstract

One of the functional vectors of the Cybersecurity Strategy of Ukraine is the development and implementation of protection systems for various information platforms in society's infrastructure, particularly focusing on creating safe technologies to detect deepfake-modifications of biometric images, based on neural networks in cyberspace. This paper presents the security principles of neural network information technologies (IT) within the context of deepfake-modifications. It delineates a basic approach for safely detecting deepfake-modifications in biometric images and outlines a security methodology for multi-level neural network IT systems, organized according to the "object – threat – protection" concept. The basic approach integrates information neural network technology with decision support IT, structured by a modular architecture for detecting deepfake modifications. This architecture operates across the stages of "pre-processing of feature data – classifier training." The core of the IT security methodology emphasizes the integrity of neural network systems for detecting deepfakemodifications in biometric images, coupled with data analysis systems that execute the information process of "dividing video files into frames – detecting and processing features – evaluating the accuracy of image classifiers. The security methodology for multi-level neural network IT relies on systemic and synergistic approaches to construct a comprehensive IT security system. This system accounts for the possibility of emergent threats and incorporates cutting-edge countermeasure technologies at both hardware and software levels. The proposed comprehensive security system for detecting deepfake-modifications in biometric images encompasses hardware and software tools across several segments: automated classifier accuracy assessment, real-time deepfake-modification detection, sequential image processing, and classification accuracy evaluation utilizing cloud computing.