iSignDB: A biometric signature database created using smartphone

iSignDB: A biometric signature database created using smartphoneSuraiya Jabin, Sumaiya Ahmad, Sarthak Mishra, and Farhana Javed ZareenDepartment of Computer Science, Jamia Millia Islamia, New Delhi-110025, IndiaIt's a database created to design a foolproof biometric signature authentication system for smartphone users. The dataset iSignDB is created to implement a novel anti-spoof biometric signature authentication for smartphone users.We named it iSignDB as we collected it using licensed MathWorks cloud account and withtwodevicesiPhone 7 Plus and Redmi Note 7 for capturing dynamic signatures.A total of 48 subjects volunteered for data collection out of which we identified 32 users as genuine signature contributors and 16 users as fake signature contributors with skilled forgery.Data was collected in 3 different sessions separated by at least 20 days in order to capture the emotional intelligence of users.During each session, one pair of subjects, out of which one subject contributed 10 original signatures and the other contributed 5 fake signatures.For obtaining a fake signature, a subject was allowed to practice not only the signature image of a genuine user but also the behaviourism (e.g. number of touchpoints, style of finger movement while signing, etc.) with genuine signer signs on the touch screen of a smartphone.A total of 30 genuine and 15 fake samples were collected for every 32 users.One sign of a user contains a sensor log of these devices captured using sensors present in device iPhone 7 Plus: Accelerometer, Gyroscope, Magnetometer, and GPS, etc along with images of signature as obtained by performing a sign on the touch screen of the device.Currently, we only put biometric sign database of one user only in this repository, but as soon as this work is published, we will make the full database of 32 users available with terms and conditions.We successfully trained 32 BiLSTM models on dynamic signature dataset created with EER of 3.35% which is a significant improvement overall such models in existing literature (HMOG, and eBioSignDS 2).We provide Matlab code (compatible with MATLAB 2020a licensed version) for training, testing, and calculating EER in this repository (https://github.com/suraiyajabin/iSignDB2020).iSignDB will be made available to other researchers only after signing its "Term of use" agreement.Nomenclature for files in the dataset iSignDB: (each sign with 4 sensor logs corresponding to Acceleration, Angular Velocity, Magnetic Field, and Position, and image of signature u01_s3_r010_AngVel.txt : means a signature of user 1, on session 3, real signature, 10th sample’s Angular velocity sensor log u01_s1_f02_MagField.txt : means a signature of user 1, on session 1, fake signature, 2nd sample’s magnetic field sensor log u01_s1_r01_im.png : image of the genuine signature of user 1, captured on session 1, sample 1

# iSignDB：一款基于智能手机的生物特征签名数据库

Suraiya Jabin、Sumaiya Ahmad、Sarthak Mishra与Farhana Javed Zareen，印度新德里贾米亚·米利亚·伊斯兰大学计算机科学系，邮编110025

本数据集专为智能手机用户研发无懈可击的生物特征签名认证系统而构建，旨在实现面向智能手机用户的新型反欺骗生物特征签名认证方案。

本数据集之所以命名为iSignDB，是因为我们通过授权的MathWorks云账户，使用iPhone 7 Plus与Redmi Note 7两台设备采集动态签名样本。

共有48名志愿者参与数据采集，最终筛选出32名真实签名样本提供者，以及16名具备熟练伪造能力的虚假签名样本提供者。数据采集分为3个独立会话，各会话间隔至少20天，以记录用户的情绪智能相关签名行为特征。

每个会话中，每对受试者里，1名受试者提供10份原始真实签名样本，另1名提供5份虚假签名样本。为采集虚假签名样本，受试者可在智能手机触摸屏上练习临摹目标真实用户的签名笔迹，同时模仿真实签名者的签名行为模式，例如触摸点数、签名时的手指移动风格等。

我们共为32名真实用户各采集了30份真实签名样本与15份虚假签名样本。每位用户的单条签名数据包含iPhone 7 Plus设备内置传感器采集的日志信息，包括加速度计（Accelerometer）、陀螺仪（Gyroscope）、磁力计（Magnetometer）、GPS等传感器数据，同时包含在设备触摸屏上签名时生成的签名图像。

当前本仓库仅上传了1名用户的生物特征签名数据库，待本研究正式发表后，我们将按照使用条款开放全部32名用户的完整数据集。

我们基于本动态签名数据集成功训练了32个双向长短期记忆网络（BiLSTM）模型，等错误率（Equal Error Rate, EER）为3.35%，相较于现有文献中同类模型（如HMOG与eBioSignDS 2）实现了显著性能提升。

本仓库（https://github.com/suraiyajabin/iSignDB2020）提供了适配MATLAB 2020a授权版本的Matlab代码，用于模型训练、测试及等错误率（EER）计算。

iSignDB仅在签署《使用条款》协议后，方可向其他研究人员开放使用。

数据集iSignDB的文件命名规范如下：
每条签名对应4份传感器日志文件（分别对应加速度、角速度、磁场与位置数据），以及一张签名图像文件。示例如下：
u01_s3_r010_AngVel.txt：表示用户1、会话3、真实签名、第10份样本的角速度传感器日志文件；
u01_s1_f02_MagField.txt：表示用户1、会话1、虚假签名、第2份样本的磁场传感器日志文件；
u01_s1_r01_im.png：表示用户1、会话1、第1份真实签名样本的图像文件。