This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Is It Really You? Exploring Biometric Verification Scenarios in Photorealistic Talking-Head Avatar Videos
Created by
Haebom
Author
Laura Pedrouzo-Rodriguez, Pedro Delgado-DeRobles, Luis F. Gomez, Ruben Tolosana, Ruben Vera-Rodriguez, Aythami Morales, Julian Fierrez
Outline
This paper focuses on the identity theft risks posed by the rise of realistic talking-head avatars. Because fraud detection based solely on avatar appearance and voice is difficult, we propose a biometric method for identity verification utilizing facial movement patterns. We construct a new dataset of realistic avatar videos generated using a state-of-the-art avatar generation model (GAGAvatar) and propose a lightweight, explainable spatiotemporal graph convolutional network (GCN) architecture that utilizes temporal attention pooling. Experimental results demonstrate that facial movement cues achieve meaningful identity verification performance with an area under the curve (AUC) approaching 80%. The proposed benchmark and biometric system are made available to the research community, highlighting the need for more advanced behavioral biometric defenses in avatar-based communication systems.
Takeaways, Limitations
•
Takeaways:
◦
A new perspective on identity theft risks in avatar-based communication systems and exploring solutions.
◦
Validation of the effectiveness of behavioral biometrics using facial movement patterns and achievement of high performance.
◦
Enabling research by providing a new avatar video dataset and a lightweight, explainable biometric system.
◦
Provides important Takeaways for enhancing the security of avatar-based systems.
•
Limitations:
◦
An AUC value of 80% is not a perfect system, and further research is needed to achieve higher accuracy.
◦
There may be limitations to the diversity and scale of the dataset (e.g., different races, ages, facial expressions, etc.).
◦
Further validation of the proposed system's application in real-world environments (e.g., changes in lighting, camera angles, etc.) is required.
◦
The system's vulnerability to complex camouflage techniques needs to be reviewed.
