Scholarlink

<p>Analyzing heart activity via remote photoplethysmography (rPPG) from facial videos holds potential for liveness detection in face recognition, as inanimate presentation attacks yield absent or degraded rPPG signals. However, the susceptibility of rPPG to environmental noise creates a vulnerability, allowing for the injection of fabricated rPPG signals that can falsely indicate liveness. In this study, we propose the Oulu Remote-photoplethysmography Presentation Attacks Database (OR-PAD), which is the first rPPG-based presentation attack dataset. OR-PAD includes 25 unique attack scenarios, simulating printed and replayed faces with forged rPPG signals generated through illumination, motion, and pixel variations. To comprehensively evaluate the impact of our proposed attack scenarios, we not only analyzed a wide range of rPPG methods, from handcrafted to deep learning (end-to-end, non-end-to-end, CNN, transformer, and self-supervised), but also directly assessed four rPPG-based face anti-spoofing techniques on our dataset. Ultimately, our findings underscore critical vulnerabilities and necessitate the development of robust security measures. Our implementation code and instructions for requesting the dataset can be found at: <a href="https://github.com/marukosan93/OR-PAD">https://github.com/marukosan93/OR-PAD</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Oulu Remote-photoplethysmography Presentation Attacks Database (OR-PAD)

  • 期刊论文

  • 发表日期：

    2025年12月28日

查看全文
  • Marko Savic,
  • Guoying Zhao

摘要

Analyzing heart activity via remote photoplethysmography (rPPG) from facial videos holds potential for liveness detection in face recognition, as inanimate presentation attacks yield absent or degraded rPPG signals. However, the susceptibility of rPPG to environmental noise creates a vulnerability, allowing for the injection of fabricated rPPG signals that can falsely indicate liveness. In this study, we propose the Oulu Remote-photoplethysmography Presentation Attacks Database (OR-PAD), which is the first rPPG-based presentation attack dataset. OR-PAD includes 25 unique attack scenarios, simulating printed and replayed faces with forged rPPG signals generated through illumination, motion, and pixel variations. To comprehensively evaluate the impact of our proposed attack scenarios, we not only analyzed a wide range of rPPG methods, from handcrafted to deep learning (end-to-end, non-end-to-end, CNN, transformer, and self-supervised), but also directly assessed four rPPG-based face anti-spoofing techniques on our dataset. Ultimately, our findings underscore critical vulnerabilities and necessitate the development of robust security measures. Our implementation code and instructions for requesting the dataset can be found at: https://github.com/marukosan93/OR-PAD.