【数据集介绍】The Idiap Research Institute REPLAY-ATTACK Database

【编辑时间】2018.09.12

【数据集介绍】The Idiap Research Institute REPLAY-ATTACK Database

 

一、数据集整体描述

 

Spoofing Attacks Description

----------------------------

The 2D face spoofing attack database consists of 1,300 video clips of photo

and video attack attempts of 50 clients, under different lighting conditions.

 

The data is split into 4 sub-groups comprising:

 

1. Training data ("train"), to be used for training your anti-spoof classifier;

2. Development data ("devel"), to be used for threshold estimation;（验证集）

3. Test data ("test"), with which to report error figures;

4. Enrollment data ("enroll"), that can be used to verify（核实） spoofing sensitivity

on face detection algorithms.

【1、训练数据（“train”），用于训练你的反欺骗分类器；

 2.开发数据（“devel”），用于阈值估计；

3.测试数据（“test”），用它来报告错误数字；

4.注册数据（“eroll”），可用于验证欺骗感兴趣的人脸检测算法。】

 

Clients that appear in one of the data sets (train, devel or test) do not

appear in any other set.

 

二、数据集具体描述

Database Description

【视频文件的获取与保存】

All videos are generated by either having a (real) client trying to access a

laptop through a built-in webcam or by displaying a photo or a video recording

of the same client for at least 9 seconds. The webcam produces colour videos

with a resolution of 320 pixels (width) by 240 pixels (height). The movies

were recorded on a Macbook laptop using the QuickTime framework (codec: Motion

JPEG) and saved into ".mov" files.  The frame rate is about 25 Hz. Besides

the native support on Apple computers, these files are *easily* readable using

mplayer, ffmpeg or any other video utilities available under Linux or MS

Windows systems.

【所有视频都是通过让一个（真正的）客户端试图通过内置的网络摄像头访问笔记本电脑或者通过显示同一客户端的照片或视频记录至少9秒来生成的。网络摄像机产生分辨率为320像素（宽度）240像素（高度）的彩色视频。影片使用QuiTime框架（编解码器：Motion  JPEG）记录在MacBook笔记本电脑上，并保存到“.MOV”文件中。帧速率约为25赫兹（每秒25帧）。除了苹果电脑的本土支持外，使用mPlayer、ffmpeg或在Linux或ms windows系统下可用的任何其他视频实用程序，这些文件都很容易可读性。】

 

【两种灯光条件】

Real client accesses as well as data collected for the attacks are taken under

two different lighting conditions:

 

* **controlled**: The office light was turned on, blinds are down,

background is homogeneous;

* **adverse**: Blinds up, more complex background, office lights are out.

 

【*可控的*：办公室的灯亮了，百叶窗掉下来，背景是均匀的；

*不利的，相反的*：窗帘，更复杂的背景，办公室灯光熄灭】

 

To produce the attacks, high-resolution photos and videos from each client were taken under the same conditions as in their authentication sessions, using a Canon PowerShot SX150 IS camera, which records both 12.1 Mpixel photographs and 720p high-definition video clips. The way to perform the attacks can be divided into two subsets: the first subset is composed of videos generated using a stand to hold the client biometry ("fixed"). For the second set, the attacker holds the device used for the attack with their own hands. In total, 20 attack videos were registered for each client, 10 for each of the attacking modes just described:

 

【攻击数据集的两个子集（两种攻击模式）】

【为了产生攻击（数据），使用佳能power.sx150 is照相机，在相同条件下给每个客户拍摄高分辨率照片和视频，该照相机记录12.1Mpixel照片和720p高清晰度视频片段。

执行攻击的方式可以分为两个子集：第一个子集是由使用机架来固定客户的生物测定生成（“固定”）的视频组成的。对于第二组，攻击者用自己的双手持有用于攻击的设备。总共为每个客户端注册了20个攻击视频，对于刚才描述的每种攻击模式，都有10个攻击视频：】

 

* 4 x mobile attacks using an iPhone 3GS screen (with resolution 480x320

  pixels) displaying:

   * 1 x mobile photo/controlled

   * 1 x mobile photo/adverse

   * 1 x mobile video/controlled

   * 1 x mobile video/adverse

* 4 x high-resolution screen attacks using an iPad (first

generation, with a screen resolution of 1024x768 pixels) displaying:

   * 1 x high-resolution photo/controlled

   * 1 x high-resolution photo/adverse

   * 1 x high-resolution video/controlled

   * 1 x high-resolution video/adverse

* 2 x hard-copy print（硬拷贝打印） attacks (produced on a Triumph-Adler DCC 2520

color laser printer) occupying the whole available printing surface on A4 paper

for the following samples:

   * 1 x high-resolution print of photo/controlled

   * 1 x high-resolution print of photo/adverse

 

【图片分配到各个数据集】

The 1300 real-accesses and attacks videos were then divided in the following

way:

 

* **Training set**: contains 60 real-accesses and 300 attacks under

different lighting conditions;（比例1:5）

 

* **Development set**: contains 60 real-accesses and 300 attacks under

different lighting conditions;

 

* **Test set**: contains 80 real-accesses and 400 attacks under

different lighting conditions;

 

* **Enrollment set**: contains 100 real-accesses under different

lighting conditions, to be used **exclusively** for studying the baseline

performance of face recognition systems.

【光照条件，专门用于研究人脸识别系统的基线性能。】

 

三、人脸定位（策略）

Face Locations

==============

We also provide face locations automatically annotated by a cascade of classifiers based on a variant of Local Binary Patterns (LBP) referred as Modified Census Transform (MCT) [Face Detection with the Modified Census Transform, Froba, B. and Ernst, A., 2004, IEEE International Conference on Automatic Face and Gesture Recognition, pp. 91-96]. The automatic face localisation procedure works in more than 99% of the total number of frames acquired. This means that less than 1% of the total set of frames for allvideos do not possess annotated faces. User algorithms must account for this fact.

 

【我们还提供基于局部二进制模式（LBP）的变体（称为修改普查变换（MCT）的分类器级联自动注释的人脸位置[[Face Detection with the Modified Census Transform]。自动人脸定位过程在所获取的帧总数的99%以上。这意味着对于所有视频的总帧集的不到1%不具有注释的脸。用户算法必须解释这个事实】

 

四、准许的生物特征处理（变换）
 

Protocol for Licit Biometric Transactions(准许的生物特征处理（变换））

-----------------------------------------

 

It is possible to measure the performance of baseline face recognition systems on the 2D Face spoofing database and evaluate how well the attacks pass such systems or how, otherwise robust they are to attacks. Here we describe how to use the available data at the enrollment set to create a background model, client models and how to perform scoring using the available data.

【可以测量二维人脸欺骗数据库上的基线人脸识别系统的性能，并评估攻击通过这些系统的程度或它们对攻击的鲁棒性。这里我们描述如何使用enrollment set数据集中的可选数据创造一个背景模型，客户模型以及如何使用这些数据进行评分】

 

1. Universal Background Model (UBM): To generate the UBM, subselect the training-set client videos from the enrollment videos. There should be 2 per client, which means you get 30 videos, each with 375 frames to create the model;

【1、通用背景模型：若要生成UBM，请从注册视频中选择培训集客户端视频。每个客户端应该有2个，这意味着你得到30个视频，每一个都有375个帧来创建模型；】

2. Client models: To generate client models, use the enrollment data for clients at the development and test groups. There should be 2 videos per client (one for each light condition) once more. At the end of the enrollment procedure, the development set must have 1 model for each of the 15 clients available in that set. Similarly, for the test set, 1 model for each of the 20 clients available;

3. For a simple baseline verification, generate scores **exhaustively** for all videos from the development and test **real-accesses** respectively, but **without** intermixing accross development and test sets. The scores generated against matched client videos and models (within the subset, i.e. development or test) should be considered true client accesses, while all others impostors;

4. If you are looking for a single number to report on the performance do the following: exclusively using the scores from the development set, tune your baseline face recognition system on the EER of the development set and use this threshold to find the HTER on the test set scores.

 

五、欺骗攻击协议

Protocols for Spoofing Attacks

------------------------------

 

Attack protocols are used to evaluate the (binary classification) performance of counter-measures to spoof attacks. The database can be split into 6 different protocols according to the type of device used to generate the attack: print, mobile (phone), high-definition (tablet), photo, video or grand test (all types). Furthermore, subsetting can be achieved on the top of the previous 6 groups by classifying attacks as performed by the attacker bare hands or using a fixed support. This classification scheme makes-up a total of 18 protocols that can be used for studying the performance of counter-measures to 2D face spoofing attacks.  The table bellow details the amount of video clips in each protocol.

【攻击协议用于评估欺骗攻击的对抗措施（二进制分类）的性能。根据生成攻击的设备类型，可以将数据库分成6种不同的协议。打印，手机（手机），高清（平板），照片，视频或大测试（所有类型）。此外，可以通过将攻击者徒手执行的攻击或使用固定支持来进行分类在前6组之上实现子集。该分类方案由18个协议组成，可用于研究二维人脸欺骗攻击的对策性能。下面表格中详细描述了每个协议中的视频剪辑的数量。】

 

Distribution of videos per attack-protocol in the 2D Face spoofing database.

【Digitalphoto---数码照片  Photo---普通照片  Grandtest：Photo+Video】

 

 

 

--------------------