Understanding iBeta certification: Eligible Datasets for Face Recognition Technology Testing

Why choose iBeta, and which products require testing?

iBeta Quality Assurance is a biometrics testing lab that works with various technology companies to ensure their products meet the highest standards. The lab is accredited by NIST NVLAP (NVLAP Testing Lab Code 200962-0) and has a Quality Management System and biometrics test procedures independently audited by NVLAP in accordance with ISO/IEC 17025:2017. iBeta is also the first biometric testing lab accredited by the FIDO Alliance for biometric evaluations in line with their Biometric Component Certification Program. This program certifies subcomponents for globally recognized standards in Presentation Attack Detection (PAD) and biometric recognition performance. iBeta received a Mastercard accreditation for its biometrics test lab, which includes testing for facial recognition, palm recognition, voice recognition, and fingerprint recognition on mobile and wearable devices.

iBeta certification is important for machine learning companies because it demonstrates that their machine learning models have been rigorously tested and evaluated for accuracy, fairness, and security. Obtaining this certification boosts client and customer trust and enhances the company’s reputation in the industry.
While iBeta certification isn’t mandatory, companies involved in the creation and application of facial recognition technology may decide to pursue iBeta certification to showcase the precision and dependability of their technology.

A high-quality dataset is crucial for achieving success in iBeta tests, as it forms the foundation for testing facial recognition technology. The efficacy and reliability of the technology tested directly depend on the dataset’s quality, making the creation and use of strong datasets imperative for successful certification of machine learning-based facial recognition systems.

“If the technology passes the testing process, iBeta will issue a certification that demonstrates that the technology has been rigorously tested and evaluated for accuracy, fairness, and security. This certification can help build trust with clients and customers and improve the reputation of the company in the industry”

Roman Kucev
CTO&Co-founder TrainingData

How to prepare a dataset for iBeta certification?

To achieve iBeta certification, companies need to submit their face recognition technology to rigorous testing using a suitable dataset. The iBeta certification process comprises two levels that differ in terms of the extent and stringency of the testing criteria.

The first step is to collect a diverse, unbiased, and representative dataset of images and photos to test the face recognition technology. The quality of the dataset directly impacts the accuracy and reliability of the technology being tested.
For Level 1 certification, the machine learning dataset must adhere to the following specific requirements:

• Each individual must have a minimum of 5 images in the dataset,
• The attacks used must be captured using a high-resolution camera, with varied backgrounds and attributes,
• The attacks must be recorded in a controlled environment with standardized lighting conditions,
• The attack material should be diverse but cost-effective (priced at no more than $30),
• The dataset should exhibit an appropriate distribution of individuals with regards to their gender and ethnicity.
• The dataset collection process must comply with the iBeta Data Collection Guidelines.

For the purpose of achieving Level 2 certification, the dataset must meet the following supplementary criteria:

• Each subject in the dataset is expected to have no fewer than 10 images.
• The attacks must be captured by a high-resolution camera, exhibiting diversity in backgrounds and attributes,
• The attacks should involve different material including latex and individual silicone masks
• The attacks must be filmed under various lighting conditions to represent real-world variations,
• The dataset must have adequate distribution of individuals based on gender and ethnicity,
• The data collection process must adhere to the iBeta Data Collection Guidelines.

“Collecting the dataset for Level 2 certification is particularly challenging. In addition to images and videos, expensive cooperative subjects and equipment are required, such as a 3D printer, resin mask, and latex mask. These types of equipment are not only costly to purchase and produce but also require a lot of attention to ensure the proper usage and capture of accurate photos.”

Yura Mayer
Product Manager TrainingData

What do we offer businesses developing face recognition products?

Collecting such a dataset is a challenging and time-consuming task. TrainingData.pro offers ready-to-use datasets for iBeta Level 1 and Level 2 that meet the certification standards.
To view examples, download the presentation with data, and receive consultation on datasets for iBeta, you can visit the Presentation Attack Detection and Liveness Detection pages.
Using our datasets can save time and resources for organizations while ensuring that their face recognition technology is of the highest quality. A thoughtfully compiled dataset provides product owners and CEOs with the assurance that their technology is precise and dependable across a variety of populations.
Apart from its importance for iBeta certification, high-quality biometric datasets are also crucial for the development and improvement of face recognition technology. As technology advances, it is crucial to continue collecting and updating datasets to ensure that the technology remains accurate and reliable. At TrainingData.pro, we offer a wide range of datasets that can help ML and CV developers train their models and achieve high scores in anti-spoofing and other biometric systems.