CUP (Columbia University Palm-vein dataset)
收藏资源简介:
CUP是首个公开的基于视频的掌静脉数据集,由哥伦比亚大学创建,旨在研究真实部署中表面噪声、光照变化等挑战性条件对识别的影响。数据集包含5,049个两秒的近红外视频片段,覆盖109名受试者在四种表面条件(清洁、温暖、潮湿、脏污)下的掌静脉图像,并附有生理和人口统计元数据。采集过程使用移动摄像头和850nm近红外发光二极管照明,在自然姿态和距离变化下录制,每个手掌均重复录制所有条件。该数据集主要服务于掌静脉识别系统在非理想环境下的鲁棒性评估与公平性分析,为算法应对真实世界退化提供了关键基准。
CUP is the first publicly released video-based palm vein dataset developed by Columbia University, designed to study the impact of challenging real-world deployment conditions such as surface noise and illumination variations on recognition systems. The dataset contains 5,049 two-second near-infrared video clips, capturing palm vein images from 109 subjects across four surface conditions: clean, warm, moist, and contaminated, with accompanying physiological and demographic metadata. The data collection process employed a mobile camera and 850nm near-infrared light-emitting diodes (LEDs) for illumination, with recordings taken under varying natural hand postures and distances. Each palm was repeatedly recorded under all four surface conditions. This dataset primarily supports robustness evaluation and fairness analysis of palm vein recognition systems in non-ideal environments, serving as a critical benchmark for algorithms to address real-world degradation.




