🧬 Biometrics & Face¶

The Core AI Technology Behind eKYC¶

This section is the technical heart of the eKYC knowledge base. It covers every aspect of face-based biometric verification — from detection and recognition to liveness, presentation attack detection, deepfakes, injection attacks, and the metrics that measure it all.

Articles in This Section¶

Face Detection & Preprocessing¶

#	Article	What You'll Learn
1	Face Detection	SCRFD, RetinaFace, BlazeFace — architectures, landmarks, speed
2	Face Alignment & Preprocessing	Affine transforms, normalization, quality assessment
3	Face Quality Assessment	ICAO compliance, blur, exposure, pose estimation

Face Recognition¶

#	Article	What You'll Learn
4	Face Recognition Overview	End-to-end pipeline, embeddings, matching
5	Face Recognition Architectures	ArcFace, CosFace, AdaFace, ElasticFace
6	Face Matching & Thresholds	Cosine similarity, threshold tuning, 1:1 vs 1:N
7	Cross-Age Face Matching	Age-invariant recognition for old ID photos
8	Cross-Quality Face Matching	Matching low-quality ID photos with HD selfies

Face Liveness & Presentation Attack Detection¶

#	Article	What You'll Learn
9	Face Liveness Detection Overview	What, why, how — complete introduction
10	Presentation Attack Types	Print, screen, mask, makeup — the attack taxonomy
11	Active vs Passive Liveness	Challenge-response vs single-frame detection
12	Liveness Model Architectures	CNN, ViT, multi-task, auxiliary supervision
13	Liveness Datasets	OULU-NPU, CASIA-FASD, SiW, CelebA-Spoof, and more
14	Domain Generalization for Liveness	SSDG, MADDG, cross-dataset generalization
15	iBeta Certification	ISO 30107-3 testing, Level 1 vs Level 2

Advanced Threats¶

#	Article	What You'll Learn
16	Deepfake Detection	GAN artifacts, temporal analysis, detection methods
17	Injection Attacks	Virtual cameras, API injection, emulator attacks
18	3D Mask Attacks	Silicone, resin, 3D-printed masks
19	Adversarial Attacks on Face Models	Perturbations, patches, physical adversarial attacks

Biometric Metrics & Standards¶

#	Article	What You'll Learn
20	Biometric Performance Metrics	FAR, FRR, EER, APCER, BPCER, ACER
21	NIST FRVT	Face Recognition Vendor Test — how it works, rankings
22	ISO 30107 — PAD Testing	The standard for presentation attack detection testing
23	Biometric Fairness & Bias	Demographic performance differentials, NIST FATE

Advanced Topics¶

#	Article	What You'll Learn
24	Face Anti-Spoofing Feature Extraction	Texture, frequency, depth, temporal features
25	Multi-Spectral Liveness	NIR, depth sensors, structured light
26	Behavioral Biometrics	Keystroke, gait, device interaction patterns
27	Voice Biometrics	Speaker verification, anti-spoofing
28	Fingerprint in eKYC	Aadhaar biometric, on-device fingerprint
29	Iris Recognition	IriCode, Aadhaar iris, mobile iris
30	Face De-Duplication (1:N)	Preventing multiple accounts with same face
31	Template Protection	Cancelable biometrics, homomorphic encryption
32	On-Device Biometric Processing	Mobile inference, ONNX, CoreML, TFLite

How These Concepts Connect¶

graph TD
    A[Camera Capture] --> B[Face Detection<br/>SCRFD, RetinaFace]
    B --> C[Face Alignment<br/>Affine transform]
    C --> D[Face Quality Check<br/>Blur, pose, exposure]

    D -->|Quality OK| E[Parallel Processing]
    D -->|Poor quality| F[Reject + retry]

    E --> G[Face Liveness / PAD<br/>Is this a live face?]
    E --> H[Face Recognition<br/>Generate embedding]

    G -->|Spoof detected| I[Reject]
    G -->|Live face| J[Continue]

    H --> K[Face Matching<br/>Compare with ID photo]
    K -->|Match > threshold| L[✅ Verified]
    K -->|Match < threshold| M[❌ Mismatch]

    style B fill:#4051B5,color:#fff
    style G fill:#e53935,color:#fff
    style H fill:#6A1B9A,color:#fff
    style L fill:#2E7D32,color:#fff

For ML Engineers

Start with Face Liveness Detection Overview and Liveness Model Architectures if you're building PAD systems. If you're focused on recognition, start with Face Recognition Architectures and Face Matching & Thresholds.