Ultrasound-Assisted Biometric Watermarking Against Face Video Deepfakes

Authors: Guodong Liu, Lei Wang, Dongxu Liu, Qianran Qiao, Meng Li, Haipeng Dai, He Huang, Guihai Chen

Published: 2026-11-09 02:55:38+00:00

Comment: Submitted to Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT)

AI Summary

This paper introduces UltraGuard, a novel biometric video watermarking method that leverages smartphone speakers and microphones to emit ultrasonic signals and capture biometric features. These features are then fused with spatio-temporal dynamic features from video content, encoded as watermarks, and embedded into videos in real time. UltraGuard aims to provide dynamic, content-aware protection against facial video deepfakes and malicious editing throughout transmission and propagation, addressing the current limitations of offline watermarking and real-time protection.

Abstract

Currently, artificial intelligence-generated content (AIGC) poses a severe threat to facial video deepfakes, and many AIGC companies employ video watermarking to mitigate unauthorized fake video usage, but its effectiveness is limited due to watermark tampering and private-model forgery. Additionally, some schemes actively add watermarks to user videos. However, current approaches focus primarily on the quality of watermark embedding in offline scenarios, and no existing work offers a solution for protecting human facial videos in real-time online applications. To fill this gap, we propose, to the best of our knowledge, a novel biometric video watermarking method, UltraGuard, for the first time. It uses the smartphone's built-in speaker and microphone to emit ultrasonic signals and capture biometric features, which are then fused with spatio-temporal dynamic features derived from the video content. The fused features are encoded as watermarks and embedded into videos in real time, enabling dynamic, content-aware watermarking throughout transmission and propagation. Extensive experimental validation demonstrates that UltraGuard can effectively defend against malicious attacks such as forgery and editing, both at the source and during propagation, proving the system's efficacy. In the future, it can be deployed on users' mobile devices to provide real-time protection for users' facial videos, thereby preventing security and privacy breaches caused by various malicious attacks.


Key findings
UltraGuard achieved an average accuracy of 99.35% against source-level attacks and 100% against propagation-level attacks, demonstrating its effectiveness in defending against various deepfake and editing scenarios. The system exhibited robust performance across different time windows, distances, angles, and smartphone types, maintaining high accuracy (average 96.46% for legitimate samples). Edge deployment optimizations significantly reduced the model's computational and memory footprint by over 82.51%, achieving real-time processing speeds suitable for commodity smartphones.
Approach
UltraGuard uses a smartphone's speaker to emit ultrasound and a microphone to capture biometric signals, which are then fused with spatio-temporal features extracted from the video content using a pre-trained ArcFace model. These fused features are encoded into a dynamic watermark, combining a static User Key and a content-dependent Face Key, and imperceptibly embedded into the video using InvisMark. The system is optimized for real-time edge deployment on smartphones through face detection, key region localization (lips), a sliding-window cache pool, and model quantization (INT8 QAT).
Datasets
A custom dataset built with 50 volunteers, including 25 hours of legitimate samples and over 135 hours of attack samples (Simple Video Replay, Random Pairing, Advanced Video Replay, Sophisticated Video Forgery, Forgery/Watermark Reuse, Editing).
Model(s)
ArcFace (pre-trained for video encoder), InvisMark (for watermark embedding and detection), SCRFD-500M (for lightweight face detection), MobileFaceNet-based backbone (for lightweight face feature encoder).
Author countries
China