Recspace matches a live selfie to the photo on a government-issued ID — with passive liveness detection that resists photos, videos, masks, and deepfakes.
Face matching and liveness detection from a single API call — because a selfie that matches but isn't live is worse than no selfie at all.
Compare a live face to the photo on a submitted ID document. Our embedding model is robust to aging (5–10 year gaps), pose variations up to 30°, lighting changes, and beard or hairstyle differences. Similarity scores are returned on a 0–1 scale with a configurable threshold — default 0.80 for KYC.
Detect spoof attempts without asking the user to blink, smile, or turn their head. Passive-only liveness feels invisible to honest users and is far harder for fraudsters to socially engineer. iBeta PAD Level 2 certified — tested against 2,000+ presentation attacks.
Signal-level analysis detects GAN artifacts, compression anomalies, and temporal inconsistencies that distinguish synthetic media from real camera capture. Continuously updated as new generation models emerge.
Search a new applicant's face against your existing customer database for duplicate-account fraud. Returns top-N matches with similarity scores — respects your data residency. Available for enterprise plans.
Presentation Attack Detection (PAD) is a moving target. Here's what our model catches, and how we test it.
Photographs printed and held in front of camera, including high-quality glossy prints and backlit displays.
Video of another person played back on a phone, tablet, or monitor.
Silicone, resin, and paper mask attacks, including high-end cosmetic prosthetics.
Face-swapped and fully generated synthetic video streamed through virtual camera drivers.
Pre-recorded video injected into the capture pipeline, bypassing the camera entirely.
One endpoint returns everything you need to make a decision — similarity score, liveness verdict, confidence, and a signed audit record.
# Verify a selfie against an ID photo curl -X POST https://api.recspace.in/v1/face/match \ -H "Authorization: Bearer $API_KEY" \ -F "id_photo=@aadhaar_face.jpg" \ -F "selfie=@live_capture.jpg" \ -F "check_liveness=true" \ -F 'threshold=0.80' # → 200 OK · 320ms { "request_id": "face_8B3A2F", "match": { "similarity": 0.968, "threshold": 0.80, "verdict": "match" }, "liveness": { "score": 0.994, "verdict": "real", "checks": { "depth": "pass", "texture": "pass", "anti_spoof": "pass" } }, "landmarks_detected": 68, "processing_time_ms": 320 }
Accuracy claims are only as good as the auditor who tested them.
Independent lab testing against Levels 1 + 2 presentation attacks — printed photos, masks, replay attacks. Zero misses required.
Benchmarked against the NIST Face Recognition Vendor Test. Top decile for 1:1 verification and demographic fairness across ethnic groups.
Meets all RBI requirements for Video-based Customer Identification Process (V-CIP) — live in production at multiple Indian banks.
We'll ship you a sandbox and a test kit. Run your own presentation attacks. If you find a bypass, we'd like to know about it.