AI & Facial Recognition Disclosure
FaceAccess uses artificial intelligence and facial recognition technology to provide identity verification services. This disclosure explains exactly how these systems work, their accuracy, limitations, and safeguards.
🤖 Transparency Commitment
FaceAccess is committed to transparent AI use. Our facial recognition systems are used exclusively for identity verification — not surveillance, profiling, tracking, or marketing. No automated decisions with legal or similarly significant effects are made solely by AI without appropriate human oversight.
1. What AI Technology We Use
FaceAccess employs the following AI-powered technologies in our biometric pipeline:
🔍 Face Detection Engine
A real-time face detection algorithm that locates and tracks faces within camera frames. It evaluates frame quality metrics including brightness (40–220 range), sharpness (Laplacian variance ≥ 15), and facial coverage (8–75% of frame area) to ensure capture quality meets minimum thresholds before proceeding.
📐 Head Pose Estimation
A geometric pose estimation model that determines the yaw (left/right), pitch (up/down), and roll orientation of the face. This powers our step-guided enrollment flow that captures facial templates from five angles: center, left turn, right turn, upward tilt, and downward tilt.
🧬 Face Embedding Generator
A 128-dimensional feature extraction model (similar in approach to ArcFace/FaceNet) that converts a 96×96 pixel normalized face crop into a floating-point mathematical vector. This vector is L2-normalized and encrypted for storage. No raw image is stored — only the mathematical representation.
👁️ Liveness Detection (Anti-Spoofing)
A multi-frame liveness analysis system that evaluates texture consistency, micro-motion patterns, and skin-coverage scoring to distinguish a live person from a printed photo, phone screen, or mask. The anti-spoof threshold is 0.72 on a 0–1 scale.
✅ Identity Matching (Verification)
Cosine similarity comparison between the captured embedding and stored enrolled templates. A match requires a similarity score exceeding configurable thresholds: High confidence ≥ 0.85, Medium ≥ 0.65, Low ≥ 0.45. Access decisions may require High confidence depending on security settings.
2. How the Enrollment Process Works
When you enroll your face:
- Your device camera activates and streams video frames at up to 25 fps
- The system guides you through 5 poses: center, turn left, turn right, look up, look down
- Each pose is held for 1–3 seconds while quality checks pass (brightness, sharpness, coverage, liveness)
- One high-quality embedding is captured per pose (5 total per enrollment)
- Embeddings are averaged or stored per-angle for multi-angle matching
- Raw video frames are immediately discarded after embedding extraction
- The final encrypted embedding set is stored in your account profile
3. How Face Verification Works
When you present your face for access:
- Camera frames are analyzed for face presence and quality
- Liveness detection is performed to reject spoofing attempts
- A face embedding is generated from the current frame
- The embedding is compared against your stored enrollment templates using cosine similarity
- If similarity exceeds the access threshold, access is granted and an event is logged
- If below threshold, access is denied and the attempt is logged
- The comparison embedding is not permanently stored
4. System Performance Metrics
5. Known Limitations and Accuracy Considerations
Like all facial recognition systems, FaceAccess AI has known limitations:
- Lighting conditions: Very low light (brightness < 40) or extreme overexposure (brightness > 220) can degrade matching accuracy. The system will prompt for better lighting.
- Occlusion: Masks, heavy glasses, or objects partially covering the face may reduce match scores. Users may need to remove occlusions for enrollment or verification.
- Camera quality: Very low resolution or heavily compressed camera feeds may reduce embedding quality and match reliability.
- Demographic performance: Facial recognition systems generally perform with varying accuracy across different demographic groups. We are committed to monitoring and improving fairness across all user populations.
- False accept / false reject: No system is perfect. Users may occasionally be incorrectly rejected (false negative) and may need to re-scan. False acceptances (incorrectly granting access) are mitigated by the anti-spoof threshold and multi-angle matching.
- Aging and appearance changes: Significant changes in appearance (major weight change, facial surgery, new facial hair) may require re-enrollment for optimal accuracy.
6. Automated Decision-Making
FaceAccess uses facial recognition to make automated access decisions (grant or deny) at configured access points. These decisions are:
- Based solely on biometric similarity scoring — not on demographic characteristics, race, gender, or other protected attributes
- Logged for audit purposes with confidence scores
- Overrideable by system administrators who can grant or deny access manually
- Subject to human review upon request — contact support@faceaccess.com
We do not use facial recognition for purposes beyond identity verification and access control. We do not use this technology for emotion recognition, demographic profiling, or any predictive analytics about individuals.
7. Data Used to Train the System
FaceAccess's current enrollment and verification pipeline is based on established open-source and commercial facial recognition research (including ArcFace, FaceNet, and related methods). We do not currently use enrolled user data to re-train or fine-tune our recognition models without separate, explicit consent.
8. No Sale or Transfer of AI Models
User biometric data and derived embeddings are never used to train third-party AI systems, never sold, and never shared with advertising or data broker platforms.
9. Human Oversight
FaceAccess maintains human oversight of its AI systems including:
- Regular review of system accuracy metrics and false-rejection rates
- Manual review capability for any contested access decision
- Monitoring for demographic bias and performance disparities
- Administrator ability to override, disable, or reconfigure AI thresholds
10. Your Rights Regarding AI Decisions
You have the right to:
- Request a human review of any access decision made by our AI system
- Obtain an explanation of why a verification attempt failed
- Re-enroll your biometric data at any time to improve matching accuracy
- Opt out of biometric authentication and use alternative access methods where available
Contact support@faceaccess.com for any AI-related rights request.
11. Contact
For questions about FaceAccess AI systems:
support@faceaccess.com