CRVR: Continuous Representation-Driven Video Frame Modulation Against rPPG Heart Rate Measurement
Keywords: Adversarial video attack, Remote physiological heart rate measurement.
TL;DR: This paper introduces CRVD, a novel framework for adversarial video attacks on rPPG heart rate measurement, utilizing continuous representation-driven resampling to subtly yet effectively disrupt detection algorithms.
Abstract: Facial video-based remote physiological measurement (rPPG) has gained prominence for its ability to non-invasively estimate vital signs such as heart rate (HR).The foundation of rPPG lies in using a camera to record facial videos at a certain frame rate, allowing the capture of rapid skin color changes necessary for HR measurement. Inspired by this property, we identified a new task, that is, to embed malicious information into facial videos by subtly modulating frames and generating frames corresponding to the modified rate. With this task, we can mislead state-of-the-art rPPG HR methods through natural and imperceptible frame modulation changes, aiming for two objectives: testing the resilience of rPPG methods against frame modulation variations and safeguarding heart rate data, which is crucial for individual privacy. However, such a task is non-trivial and should be capable of automatically adapting to different input videos and generating natural, imperceptible frame modulation perturbations along with frames corresponding to the modified rate. To address these challenges, we propose Continuous Representation-driven Video Resampling (CRVR), which targets precise manipulation of frame timing to subtly skew perceived HR measurements. Specifically, the CRVR method consists of two modules: Variable Frame Rate Video Resampling (VFRVR), which automatically determines the optimal resampling strategy for each frame, and Continuous Video Frame Generation (CVFG), which generates frames corresponding to the modified rate and seamlessly injects them back into the video. Extensive testing on UBFC-rPPG and PURE datasets reveals that our CRVR method successfully produces realistic, imperceptible adversarial videos that effectively mislead three different rPPG-based heart rate detection technologies.
Primary Area: other topics in machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8560
Loading