Successful iBeta certification Level 1 & 2
Task
The Data Science team from a financial institution in Latin America approached us with the task of improving their existing liveness detection model and achieving iBeta Level 1 and Level 2 certifications. Their current baseline model needed significant optimization to meet the strict certification requirements and perform effectively in real-world production environments. The company needed a reliable solution for passive liveness detection that would ensure high accuracy and resilience against various types of attacks.
Solution
We provided the client with pre-recorded datasets covering both common attacks occur in production environments and specific attacks required for iBeta certification.
For iBeta level 1 certification, we focused on 2D attacks, providing a wide range of attacks to ensure the model works well and get reliable results:
For iBeta level 2 certification, we provided video recordings of attacks using silicon mask attack dataset with 18 different masks.
Since the client used passive liveness detection, no additional data customization was required. All necessary datasets were available. Our “Zoom in / Zoom out” methodology provided high frame variability, allowing us to offer the most cost-effective solution on the market per frame basis.
Result
- Certifications Achieved: The client successfully obtained iBeta1 and iBeta2 certifications on the first attempt in less than 3 months
- Production-Ready Solution: The improved model performs well in both certification and real-world production environments
- Cost Savings: Using our pre-recorded datasets allowed the client to avoid the high costs of custom data collection and model retraining
- Improved spoof detection: The client was able to significantly improve spoof detection in their system
