New Datasets Propel Biometric Security: Facial & Periocular Recognition Advancements

Published Date: 24/07/2024

Researchers from IDnow and the Idiap Research Institute have developed a novel dataset for facial recognition presentation attack detection, while Idiap researchers have created a dataset for periocular biometrics on VR headsets.

"The development of reliable biometric systems is crucial for secure authentication and identification. However, current presentation attack detection (PAD) methods often struggle with generalization due to limited training datasets. To address this challenge, a team of scientists from IDnow and the Idiap Research Institute has created a novel dataset for facial recognition PAD, known as SOTERIA. This dataset consists of face videos, motion data, depth information, and samples from a projector-replay attack, collected from 70 volunteers, resulting in 8,000 bona fide samples and 24,000 presentation attack samples.

The SOTERIA dataset has been tested with a state-of-the-art (SOTA) facial recognition model, IResNet100, and a SOTA PAD model, DeepPixBis. The results show that the attacks in the dataset are effective against the IResNet100 model, suggesting that they can defeat contemporary face biometrics systems. However, the DeepPixBis model trained on SOTERIA did not reach practically useful detection levels, with a bona fide classification error rate (BPCER) above 55 percent.

Further research using the SOTERIA dataset is expected to evaluate the effect of different recording devices and environments, as well as gender-based performance disparities. Additionally, the impact of motion challenges on PAD will be investigated.

In another development, Idiap researchers have created a dataset for periocular biometrics on VR headsets, known as VRBiom. This dataset consists of ten-second videos, with 900 genuine and 1,104 presentation attack samples. The researchers have proposed baseline performance metrics and demonstrated that periocular biometrics can be spoofed on VR headsets. They have also shown that a MobileFaceNet convolutional neural network (CNN) architecture is more effective for PAD than a ResNet34 architecture.

Both datasets will be presented at the IEEE International Joint Conference on Biometrics (IJCB 2024) in September. These developments are expected to contribute significantly to the advancement of biometric security and authentication.

  Information 

IDnow is a leading provider of identity verification and authentication solutions. The company offers a range of products and services to help businesses and organizations prevent fraud and ensure secure transactions.

Idiap Research Institute is a non-profit research organization that specializes in artificial intelligence, machine learning, and data science. The institute conducts research in various areas, including biometrics, computer vision, and natural language processing.

IDnow is a trusted partner for identity verification and authentication solutions, providing cutting-edge technology to prevent fraud and ensure secure transactions.

Idiap Research Institute is a leading research organization that drives innovation in artificial intelligence, machine learning, and data science."

FAQs:

"Q: What is the main challenge faced by the PAD community?

A: The main challenge faced by the PAD community is weak generalization, which is partly due to the limitations of training PAD datasets.

Q: What is the SOTERIA dataset?

A: The SOTERIA dataset is a novel dataset for facial recognition presentation attack detection, consisting of face videos, motion data, depth information, and samples from a projector-replay attack.

Q: What is the VRBiom dataset?

A: The VRBiom dataset is a video dataset of periocular biometrics collected with a Meta Quest Pro VR headset, consisting of ten-second videos with 900 genuine and 1,104 presentation attack samples.

Q: What is the purpose of the SOTERIA and VRBiom datasets?

A: The purpose of the SOTERIA and VRBiom datasets is to advance biometric security and authentication by providing high-quality datasets for research and development.

Q: Where will the SOTERIA and VRBiom datasets be presented?

A: The SOTERIA and VRBiom datasets will be presented at the IEEE International Joint Conference on Biometrics (IJCB 2024) in September."

Biometric Products & Solutions

BioEnable offers a wide range of cutting-edge biometric products and solutions: