移动与桌面平台上的连续人脸认证：一项比较研究
Continuous Face Authentication on Mobile and Desktop Platforms: A Comparative Study
AI总结：
该研究对比移动与桌面平台的连续人脸认证，采用基于InsightFace的带时间信任衰减方法，发现日常使用的人脸可见性问题更影响认证性能。
AI中文摘要：
个人设备存储敏感数据并提供对敏感服务的访问权限。传统个人设备认证仅在授予访问权限时验证用户身份，若用户停止使用设备却未锁定，或被他人接管，已解锁的设备可能被未授权人员访问。连续认证可解决这一问题。本文研究设备类型和使用条件如何影响基于InsightFace且具有时间信任衰减的连续移动人脸认证方法，使用包含不同头部方向和光照条件的移动与桌面录制数据，以及无预定义任务的日常移动设备使用录制数据评估该方法。结果显示，仅设备类型影响很小，而不同使用条件确实会影响认证性能；日常移动设备使用通常对连续人脸认证更具挑战性，人脸可见性降低（包括遮挡和人脸超出相机视口）是导致错误拒绝的主要因素。
英文摘要：
Personal devices hold sensitive data and provide access to sensitive services. Conventional personal device authentication verifies users' identity only at the moment access is granted. An unlocked device may be accessed by an unauthorized person if the user stops using the device without locking it, or if another person takes over. Continuous authentication addresses this gap. This paper investigates how device type and usage conditions influence continuous mobile face authentication with an InsightFace-based approach with temporal trust decay. We evaluate the approach with mobile and desktop recordings with different head directions and lighting conditions. We also evaluate recordings from everyday mobile device use without predefined tasks. The results show that device type alone has little impact, while different usage conditions do have impact on the authentication performance. Results also show that everyday mobile device use is in general more challenging for continuous face authentication, where reduced face visibility, including occlusions and faces outside the camera viewport, is a main contributor to false rejections.
