Use Case: Image Collection Through Our Community-Driven Approach
Objective: A company specializing in KYC (Know Your Customer) systems needed to improve the accuracy of their liveliness detection system, which needs a user's selfie to verify identity. They required support with three critical data types: (1) selfies taken in a natural environment, (2) selfies with a picture attack, and (3) selfies with a rubber mask attack. We addressed this need by leveraging our community and mobile app to collect and deliver high-quality, diverse data.
How We Helped
Community-Driven Data Collection:
Using our mobile app, we engaged participants from our community to contribute selfies in the 3 scenarios mentionned above. Users were guided through simple steps to ensure clear and usable images. This approach allowed us to gather a wide range of images under varied lighting and environmental conditions, reflecting real-world scenarios.
Selfies with a picture attack: We instructed participants to simulate picture attack conditions by submitting printed photos, digital screenshots, or edited images of real individuals attempting to bypass selfie verification systems. This enabled the client to train their systems to recognize and counteract spoofing attempts, enhancing robustness against fraudulent access in real-world scenarios.
Rubber Mask Attack Simulation: Our global community simulated rubber mask attacks by wearing hyper-realistic masks designed to mimic human facial features, challenging the selfie-based verification system to detect such spoofing attempts. This diversity of mask types, ethnicities, and facial features provided a comprehensive dataset to enhance the system's global accuracy and applicability.
Secure Data Handling: All data was collected and managed through a secure, encrypted platform to ensure user privacy and compliance with data protection standards. Participants were informed of how their data would be used, fostering transparency and trust.
Efficient Delivery: The data was processed, categorized, and delivered in a structured format. This included metadata tagging for each scenario, gender, age category, ensuring seamless integration into the client's training workflows.
Benefits
High-Quality, Real-World Data: By leveraging our community, we sourced authentic, diverse data reflecting real-world conditions, improving model performance for selfies and ID verification.
Cost and Time Efficiency: Our scalable approach provided the required data faster and more affordably than traditional collection methods, such as in-house or outsourced data generation.
Global Diversity: Data from multiple countries enabled the KYC system to handle various ID formats and languages, enhancing its international usability.
Robustness in Handling Damaged IDs: Training on images of damaged IDs allowed the client to improve their system’s accuracy in verifying imperfect or worn documents.
Data Security: Our secure platform ensured compliance with privacy regulations, protecting user data and maintaining trust.
Outcome
We delivered a comprehensive, diverse dataset that allowed the KYC system to improve liveliness detection and ID verification accuracy. This collaboration empowered the client to offer a more reliable and robust solution for identity verification, demonstrating the value of our community-driven data collection approach in addressing complex AI and machine learning needs.
