Tencent Cloud Huiyan High-Availability Architecture Design for Face Identity Verification

This article introduces Tencent Cloud Huiyan, a comprehensive identity verification service that provides document OCR, liveness detection, face comparison, and multi-factor verification capabilities. During the COVID-19 pandemic, this system supported health code verification across multiple Chinese provinces, with the Shenzhen public security system alone accumulating over 6 billion visits within one month of launch.

The article details four types of liveness detection technologies:

1. Action Liveness Detection: Generates random action sequences (mouth opening, eye blinking, head shaking/nodding) that users must perform within 2-3 seconds. Attack success rate is less than 0.01%, with genuine user pass rate exceeding 99%. The system detects facial wrinkles (eye corner, mouth corner) to verify liveness.

2. Digital Liveness Detection: Uses a multi-dimensional approach combining lip movement, voice, and screen capture detection. Users read randomly generated numbers while the algorithm analyzes lip shape matching. Attackers cannot predict the random numbers, effectively defending against static photo and replay attacks.

3. Silent Liveness Detection: Requires no user interaction, completing verification in 2-3 seconds with high pass rates. Detects screen moiré patterns and edge characteristics through extensive training on live and non-live samples.

4. Light Liveness Detection: Completes face scanning within 2 seconds without interaction. The screen emits random light signals while capturing images; different wavelengths照射 face to verify three-dimensional shape and texture. Based on diffuse reflection modeling, algorithms extract facial normal vector information to reconstruct depth maps.

The architecture evolved through multiple stages: from a simple demo with direct engine calls, to introducing an Adapter layer for protocol conversion, to creating a unified Engine Middle Platform that handles scheduling, distribution, parameter configuration, and effect fusion. The final architecture includes: Access Layer (Tencent Cloud API 3.0), Business Middle Platform (image processing, video processing, billing), Data Middle Platform (statistics, reporting, quality analysis), and Engine Middle Platform connecting various AI labs (Tencent YouTu, AILab, XLab).

Key architectural design principles include: scalability with horizontal expansion at each layer, layered design with horizontal/vertical decomposition and decoupling, fault tolerance through load balancing, service degradation, dynamic scaling, and resource isolation. For handling burst traffic, the system implements pre-announced capacity planning and real-time monitoring for unexpected traffic spikes, with overload protection at both engine and logic layers to prevent cascading failures.