RECOD Mobile Presentation-Attack Dataset (RECOD-MPAD)

============= Introduction The RECOD Mobile Presentation-Attack Dataset (RECOD-MPAD) is intended for the study of presentation attacks (PAs, also known as spoof attempts) to facial recognition systems in mobile devices. It consists of frames depicting genuine attempts of unlocking a smartphone, as well as two types of presentation attacks: using printouts of the user face; or using electronic displays showing the user's face. More details can be found in the accompanying paper: Detecting face presentation attacks in mobile devices with a patch-based CNN and a sensor-aware loss function Waldir R. Almeida, Fernanda A. Andaló, Rafael Padilha, Gabriel Bertocco, William Dias, Ricardo da S. Torres, Jacques Wainer, Anderson Rocha PLoS ONE, 2020 (https://doi.org/10.1371/journal.pone.0238058) This dataset was developed as part of a research project at the RECOD lab of the Institute of Computing, University of Campinas, Brazil. ========== Statistics Number of users: 45 Men: 30/45 Glasses: 14/45 Beard: 13/45 Age range: 18-50 ============================= Basic metadata and file names Name format: <device>_<session>_<user>_<label>_<frame> Label: 0: real, genuine access attempt 1: printout attack - recaptured indoors 2: printout attack - recaptured outdoors (more light) 3: screen attack - large display (CCE TV) 4: screen attack - medium-sized display (HP monitor) Device: 1: motog3 2: xt1572 Session: 1: outdoors natural direct light 2: outdoors natural diffuse light/shadow 3: indoors, top main light 4: indoors, side light (sunlight coming through window or door) 5: indoors, low-light / noisy Example: 1_5_20_00_0050 ====================== Additional information In each sequence, the volunteers followed the same instructions: - Hold the phone as if using it normally, but keep close-to-frontal viewing angles - Rotate slowly (to change lighting and background)============= 引言 RECOD 移动端演示攻击数据集（RECOD Mobile Presentation-Attack Dataset, RECOD-MPAD）旨在研究移动设备人脸识别系统所面临的演示攻击（presentation attacks, PAs，亦称欺骗攻击）。该数据集包含记录智能手机真实解锁尝试的帧图像，以及两类演示攻击场景：使用用户面部打印件发起攻击，或使用显示用户面部的电子显示屏发起攻击。更多研究细节可参阅配套论文：《基于patch的卷积神经网络与传感器感知损失函数的移动设备人脸识别演示攻击检测》（Detecting face presentation attacks in mobile devices with a patch-based CNN and a sensor-aware loss function），作者为Waldir R. Almeida、Fernanda A. Andaló、Rafael Padilha、Gabriel Bertocco、William Dias、Ricardo da S. Torres、Jacques Wainer、Anderson Rocha，发表于PLoS ONE，2020年（链接：https://doi.org/10.1371/journal.pone.0238058）。本数据集由巴西坎皮纳斯大学计算机学院RECOD实验室的一项研究项目研发而成。 ============= 统计信息 参与者总人数：45名 男性：30/45 佩戴眼镜者：14/45 留胡须者：13/45 年龄范围：18至50岁 ============================= 基础元数据与文件名规则 文件名格式：<设备>_<会话>_<参与者>_<标签>_<帧序号> 标签说明： 0：真实合法的解锁尝试 1：纸质打印件攻击——室内翻拍场景 2：纸质打印件攻击——室外翻拍场景（光线更充足） 3：电子显示屏攻击——大型显示屏（CCE 电视） 4：电子显示屏攻击——中型显示屏（HP 显示器） 设备编号说明： 1：motog3 2：xt1572 会话场景说明： 1：室外自然直射光环境 2：室外自然漫射光/阴影环境 3：室内顶部主光源环境 4：室内侧光环境（自然光透过窗户或门照射） 5：室内低光照/高噪声环境 示例文件名：1_5_20_00_0050 ====================== 附加信息 在每个采集序列中，受试者均遵循统一操作要求： - 以日常使用手机的姿势握持设备，并保持接近正面的拍摄视角 - 缓慢转动设备（以改变光照条件与背景环境）