FrePhys: Frequency-aware Diffusion Model for Remote Physiological Measurement

Wei Qian, Dan Guo, Jinxing Zhou, Bochao Zou, Zitong YU, Meng Wang

Keywords: remote physiological measurement

TL;DR: A novel frequency-aware diffusion model for remote physiological measurement

Abstract: Remote photoplethysmography (rPPG) enables non-contact physiological monitoring by capturing subtle skin color variations in facial videos. Existing approaches predominantly rely on time-domain modeling to extract cardiac-related periodic signals, but they are highly vulnerable to motion artifacts and illumination changes, where physiological clues are easily obscured by noise. To address these challenges, we propose a Frequency-aware Physiological diffusion model, dubbed FrePhys, that integrates physiological frequency priors into rPPG estimation. Specifically, it first employs a \textit{physiological bandpass filter} to suppress out-of-band noise, followed by \textit{physiological spectrum modulation} and \textit{adaptive spectrum selection} for in-band noise suppression and pulse-related clues enhancement. A \textit{cross-domain representation learning} module then fuses frequency-domain insights with the deep time-domain features to capture spatial–temporal dependencies. Finally, a frequency-aware conditional diffusion process iteratively reconstructs high-fidelity rPPG signals. Extensive experiments on multiple datasets demonstrate that our method significantly outperforms existing state-of-the-art methods, particularly under challenging motion conditions, highlighting the effectiveness of incorporating frequency priors. The source code is available at \url{https://anonymous.4open.science/r/FrePhys}.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 517