Selfie with ID

One Shot,
Three Verifications

A single selfie-with-ID capture simultaneously verifies document authenticity, face-to-document matching, and liveness — the trifecta of identity proof that eliminates impersonation entirely.

3-in-1

Verification

98.6%

Match Accuracy

Total Time

Face Detected • Liveness Passed

ID Document Detected

Matching Face ↔ ID Photo...

Triple Verification

Three Checks, One Capture

Every selfie-with-ID frame undergoes three independent verification layers simultaneously.

Layer 1: Document Authenticity

The ID card visible in the selfie frame is validated using the same 7-stage document verification pipeline. MRZ is extracted, security features are checked, and the document photo is isolated.

Cross-reference: All extracted fields are compared against previously captured document data for consistency. Any mismatch triggers rejection.

Layer 2: Face-to-Document Match

The live selfie face is compared against the photo on the ID card using ArcFace embedding vectors. This confirms the person holding the ID is the same person pictured on it.

Threshold: Cosine similarity ≥ 0.68 required for match. At this threshold, FAR = 0.001% while FRR = 1.4%.

Layer 3: Simultaneous Liveness

While the face match is running, passive liveness checks analyze the selfie face for texture anomalies, depth consistency, and temporal coherence. The person must be physically present.

Integration: If either document validation OR face match OR liveness fails, the entire capture is rejected. All three must pass independently.

Spatial Intelligence

Frame Analysis Engine

Our ML model understands the spatial relationship between face and document in 3D space.

Dual Region Detection

Face detection: MediaPipe Face Mesh (468 landmarks)
Document detection: YOLO-Tiny custom-trained
Both regions must be present simultaneously
Minimum size thresholds prevent distant captures
Overlap detection prevents obscured documents

Photo Extraction

ID card photo region isolated via template matching
Perspective-corrected to frontal view
Color-normalized for consistent comparison
Resolution enhanced (super-resolution 2x)
Lighting equalized to match selfie conditions

Cross-Verification

Selfie face → 512-dim ArcFace embedding
ID photo → 512-dim ArcFace embedding
Cosine similarity computed in real-time
Age progression tolerance (up to 10 years)
Glasses/beard/hijab compensation layers

Fraud Detection

Edge analysis: photo-of-a-photo detection
Screen replay: moiré pattern scanning
Document swap: mid-capture switching blocked
Face swap: deepfake artifact detection
Multiple faces: rejects if > 1 face detected

Competitive Edge

Beyond Simple Photo Uploads

Most platforms ask users to separately upload a selfie and a document photo. We do it all in one intelligent capture.

PollGuard Selfie-ID

Single capture — face + document in one frame

Simultaneous liveness during capture

Auto-capture when both regions are optimal

Real-time face-to-ID matching on device

Prevents document swapping mid-stream

4-second total verification time

Age, glasses, and appearance change tolerance

Typical Approach

Two separate uploads (selfie + document)

No liveness during selfie capture

Manual shutter button with retries

Server-side matching (network required)

No temporal binding between captures

15-30 second multi-step process

Rigid matching — high reject rates

Performance

Measured Results

From 340,000+ real-world selfie-with-ID verifications.

98.6%

Face-to-ID Match Rate

Genuine document holders

96.2%

First-Attempt Success

No retries needed

4.1s

Average Verification

All 3 layers combined

0.001%

False Accept Rate

At cosine threshold 0.68

One Shot,Three Verifications