Silicone Mask dataset

There are >7K videos with 3D Silicone mask attacks tailored for iBeta level 2 certification

Check samples on Kaggle

Successfull Spoofing attack on a Liveness test by Doubango

Introduction

The Silicone Mask Biometric Attack Dataset offers a robust solution for enhancing security in liveness detection systems by simulating 3D silicone mask attacks. This dataset is invaluable for assessing and fine-tuning Passive Liveness Detection models, an essential step toward achieving iBeta Level 2 certification. By integrating realistic presentation attacks, this dataset significantly supports advancements in detecting biometric spoofing

Dataset summary

The dataset includes a comprehensive collection of approximately 10,000 videos featuring spoofing attempts with high-detail silicone masks. Designed to reflect real-world scenarios, the dataset contains various mask types and diverse settings to mimic a broad range of conditions, making it highly effective for model training and testing

Source and collection methodology

The videos capture realistic spoofing conditions using three different recording devices (modern iPhone, Xiaomi, and Samsung) to represent varied environments and devices. Additionally, the dataset simulates common interactions like head movements and blinking, adding to its effectiveness in active liveness detection

Use cases and applications

Designed for organizations aiming to meet iBeta Level 2 certification standards, this dataset supports the development of advanced anti-spoofing solutions. It’s especially beneficial for training machine learning (ML) models and facial recognition systems, enabling them to detect and respond accurately to spoofing attempts through the use of silicone masks. Key applications include improving algorithm performance for biometric authentication in security systems

Dataset features

Variety of Masks: Nine distinct silicone masks featuring diverse demographics (both male and female, covering Caucasian and Asian ethnicities).
Comprehensive Video Collection: Nearly 10,000 videos illustrating different spoofing scenarios.
Active Liveness Indicators: Includes natural head movements and blinking for enhanced training scenarios.
Multiple Angles: Five shooting angles, including close-up and distant front-facing, side views, and selfie mode, allowing multi-perspective analysis.
Device Variety: Recordings from iPhone, Xiaomi, and Samsung devices to ensure compatibility across major technologies.
Environmental Diversity: Videos captured in various lighting and background conditions to improve model robustness.
Customization Potential: Available in both RGB and USB camera inputs to support diverse research and testing needs.

File format and accessibility

Dataset details

The Silicone Mask Attack Dataset is designed to address security challenges in liveness detection systems through 3D silicone mask attacks. These presentation attacks are key for testing Liveness Detection at iBeta level 2 certification. The dataset significantly enhances the capabilities of liveness detection models

To ensure high variety, 8 silicone masks were created and used with different hair styles, clothes, face masks and glasses to shoot various attacks. Each attack is approximately 8 seconds long and includes head movements as well as blinking. The dataset is suited for both Active and Passive liveness detection

Dataset description

8 unique silicone masks (male and female, Caucasian and Asian ethnicity)
Roughly 7,000 videos with detailed spoofing detection scenarios
Head movements and blinking are included in the dataset for active liveness detection
Variations of shooitng angles, 5 angles in total: Selfie, back camera far / close, two side shots
Three different devices on iOS and Android (Modern iPhone, Xiaomi, Samsung Galaxy)
Diverse lighting and background settings to ensure robustness
Advanced set of attributes including different hair styles, glasses, and clothes

Best used for:

Potential customisation options:

Download information

A sample version of this dataset is available on Kaggle. Leave a request for additional samples in the form below

Contact us

Tell us about yourself, and get access to free samples of the dataset