CodePhys: Robust Video-Based Remote Physiological Measurement Through Latent Codebook Querying.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Signal Processing

Background:

• Remote photoplethysmography (rPPG) measures physiological signals from facial videos.
• Real-world interference like noise and motion blur degrades rPPG signal quality.
• Existing methods struggle with noisy rPPG signal recovery.

Purpose of the Study:

• To develop a robust rPPG measurement method resilient to real-world interference.
• To improve the fidelity of extracted rPPG signals from degraded facial videos.
• To introduce a novel codebook-based approach for rPPG signal estimation.

Main Methods:

• Proposed CodePhys method treats rPPG measurement as a code query task.
• Constructed a noise-free proxy space (codebook) using ground-truth PPG signals.
• Employed a spatial-aware encoder with attention and distillation loss to handle interference.

Main Results:

• CodePhys generates high-fidelity rPPG features by matching noisy features to noise-free codebook entries.
• The method effectively reduces the influence of non-periodic visual interference.
• Demonstrated superior performance over state-of-the-art methods on four benchmark datasets.

Conclusions:

• CodePhys offers a significant advancement in non-contact physiological signal measurement.
• The codebook-based approach provides robust rPPG extraction in challenging conditions.
• CodePhys achieves state-of-the-art results in both intra-dataset and cross-dataset evaluations.