In an era where high-quality photos are freely available on social media and deepfake technology is increasingly sophisticated, how can businesses be certain that the person completing an identity verification is actually present? The answer lies in AI-powered liveness detection—a critical component of modern eKYC systems.
The Growing Threat of Presentation Attacks
Presentation attacks—attempts to fool biometric systems using photos, videos, masks, or deepfakes—have become increasingly common as remote identity verification has grown. Fraudsters have evolved from simple printed photos to sophisticated attacks that can deceive basic facial recognition systems.
- 2D Photo Attacks: Holding up a printed photo or displaying one on a screen
- Video Replay Attacks: Playing a pre-recorded video of the victim
- 3D Mask Attacks: Using realistic silicone or 3D-printed masks
- Deepfake Attacks: AI-generated synthetic faces or face-swapped videos
How Liveness Detection Works
Modern liveness detection systems employ multiple techniques to ensure a real, live person is present. These can be broadly categorized into active and passive approaches, each with their own strengths and use cases.
Active Liveness Detection
Active liveness requires the user to perform specific actions, creating a challenge-response mechanism. Because the challenge is randomized and unknown in advance, it is computationally impractical for an attacker to pre-generate a matching deepfake or video replay. Assurique uses active liveness with:
- Blink Detection: Users are prompted to blink naturally — natural eyelid movement is virtually impossible to replicate convincingly with static or screen-displayed images
- Head Turn Challenges: Users turn their head in a prompted direction to verify genuine 3D presence and depth
- Challenge Randomization: The specific sequence varies per session, preventing replay attacks using pre-recorded responses
- Sub-3 Second Processing: The full active liveness sequence processes in under 3 seconds, keeping friction minimal
"Active liveness provides strong security because the specific challenge is unknown in advance—an attacker cannot pre-record the correct response."
Passive Liveness Detection
Passive liveness analyzes the video stream without requiring specific user actions:
- Micro-Movements: Natural involuntary movements that occur in live faces
- Skin Texture Analysis: Real skin has different reflective properties than photos or screens
- Depth Estimation: AI models that can infer 3D structure from 2D images
- Moiré Pattern Detection: Identifies screen-captured images by detecting display patterns
The AI Behind the Technology
Modern liveness detection relies on proprietary deep learning models trained on massive datasets of both genuine users and attack attempts. These models analyze facial images and video frames to extract features that distinguish real faces from spoofs — detecting subtle texture differences, micro-movements, and inconsistencies invisible to the human eye. Assurique's liveness engine processes a full challenge-response sequence in under 3 seconds, with active challenges including blink detection and head turn verification.
Liveness as Part of Six-Check Verification
Liveness detection — the physical presence check — is one of six security checks performed in every Assurique verification: (1) document authenticity, (2) document validity, (3) image freshness (EXIF analysis to detect reused captures), (4) physical presence via active liveness, (5) identity match via biometric face comparison, and (6) chip integrity for NFC verifications. Each check can trigger a hard gate if failed — a liveness failure results in an immediate decline, independent of the document or score. This layered approach is what separates robust eKYC from simple photo upload verification.
Implementation Best Practices
For businesses implementing liveness detection in Algeria, consider these recommendations:
- Balance Security and UX: Overly complex challenges frustrate legitimate users. Start with passive detection and escalate only when needed.
- Device Diversity: Test across different smartphone models and lighting conditions common in Algeria.
- Localization: Provide instructions in Arabic and French with clear visual guidance.
- Fail Gracefully: When liveness fails, offer alternative verification paths rather than blocking users entirely.
Key Takeaways
- Liveness detection prevents photo, video, and deepfake attacks
- Active methods use challenges; passive methods use AI analysis
- Modern systems combine multiple techniques for robustness
- Balance security requirements with user experience
- Regular updates are essential as attack methods evolve