Synthetic Face Generation for Real-Time Online Identity Protection
Аннотация
The rapid expansion of internet-based video communication has transformed personal and professional interaction but has also exposed users to serious privacy threats, particularly involving facial biometric data. Existing artificial face-generation tools primarily focus on offline image synthesis and fail to offer real-time adaptability for live communication environments. To address this limitation, this paper introduces Dynamic Context-Aware Synthetic Face Proxy (DyCASP) - a real-time privacy-preserving system that generates and animates synthetic faces dynamically during video interactions. DyCASP employs a hybrid generative model combining lightweight Generative Adversarial Networks (GANs) with neural rendering to achieve high visual realism, natural expression, and low-latency performance. A key innovation is the context-awareness module, which adapts facial lighting, expression, and environmental response based on ambient cues and conversational tone, ensuring natural and lifelike communication. DyCASP also integrates latent-space obfuscation to distort biometric identifiers, effectively countering facial recognition and spoofing attacks while preserving visual integrity. The system provides users with an intuitive interface for real-time demographic and identity customization, enabling personalized privacy control. Implemented as a hybrid edge-cloud architecture, DyCASP maintains sub-100ms response latency on consumer-grade hardware. Experimental evaluations demonstrate significant improvements in realism, privacy, and responsiveness compared to conventional offline generators. User studies further validate its acceptability and expressive naturalness in real-time communication scenarios. DyCASP thus establishes a robust foundation for next-generation, privacy-first visual interaction systems that preserve human expressiveness without compromising identity security.