Face Engine Logo

Face Engine

ArcFace 512-D • SCRFD Detection • Microsoft SQL Server
SQL Server Online
🛡️ 2FA Active
0.0 FPS 0 ms Latency 0 Face(s)
Biometric Logo Prompt

Webcam Feed Inactive

Connect your Logitech webcam and click "Start Camera" above to begin real-time face detection & ArcFace recognition.

(Recommended: 0.50 - 0.60 for ArcFace 512D)

Live Recognition Status

Logo Watermark

No Active Recognition

Faces detected in camera frame will be compared in real-time against SQL Server embeddings.

Recent Match Events

Photo Identification Service

Upload a face portrait or group image to detect identities against the Microsoft SQL Server database.

Computer File / Clipboard

Drag & Drop Face Photo Here

Supports JPG, PNG, WebP up to 20MB • Single or Multi-Face Group Photos

(ArcFace Cosine: 0.50 - 0.60 standard)
Detection: SCRFD 10G
Extractor: ArcFace ResNet-50 512-D
Database: SQL Server InsightFaceDB

No Photo Analyzed Yet

Upload any employee badge photo, surveillance still, or portrait on the left panel to execute instant biometric identification.

✓ Multi-Face Detection: Automatically tags all individuals in a group photo.
✓ Probe vs. Gallery Crop: Visual side-by-side comparison of detected face against enrolled avatar.
✓ Demographics & Quality: Estimates gender, approximate age, and face crop resolution.
✓ One-Click Quick Enroll: Unregistered detected faces can be immediately routed to registration.

Face Identification from Camera

Capture a single face photo via camera, review the frame, and click Identify Face to query SQL Server.

Camera Snapshot Mode
Camera Logo

Camera Inactive

Click "Start Camera" above to activate the camera viewfinder for snapshot identification.

(ArcFace 512D: 0.50 - 0.60 standard)
ℹ️ Click "Start Camera" above to begin. Position your face in the frame and snap.

No Camera Photo Identified Yet

Capture a face photo using the camera on the left, review the still portrait, and click Identify Face to find matching records in the database.

✓ Controlled Single Capture: No continuous stream clutter — you control exactly when to identify.
✓ Review Before Query: Inspect the captured still to ensure clarity and lighting before submitting.
✓ Probe vs. Enrolled Comparison: Side-by-side visual comparison with matching confidence score and Barcode ID.
✓ Instant One-Click Route: If no match is found, easily send the captured image directly to registration.

Facial Image Capture

Capture a clean, frontal photo via Logitech webcam or upload an image file.

Position Face Here
ℹ️ Webcam ready. Click "Snap from Webcam" to capture.

Registrant Information

Enter user details to bind to the 512-D ArcFace biometric embedding.

Unique Barcode / Alphanumeric identifier stored in SQL Server.

🛡️ Biometric Security Safeguards

  • Extracts 512-dimensional normalized ArcFace representation via DeepInsight buffalo_l.
  • Raw vector stored in Microsoft SQL Server VARBINARY(MAX) with parameterized queries.
  • Single-face constraint enforced to prevent identity contamination.

Batch Face Registration Engine

High-throughput automated enrollment for 500+ identities. Import directly from a local disk folder or browser upload with automatic face detection, ArcFace feature extraction, biometric duplicate checks, and fault-tolerant SQL Server persistence.

Folder Source & Extraction Configuration

Specify a directory path on the local system or select a folder via browser upload.

Drag & Drop Image Folder Here Supports JPEG, PNG, WEBP, BMP up to thousands of files
Determines how Person ID and Full Name are parsed from filenames.
Behavior when Person ID is already registered in SQL Server.
Action to take when an image contains more than one detected face.
Thread pool concurrency for ONNX inference and DB write.
Rejects blurry, occluded, or low-confidence face detections.

Face Image A (Probe)

Click or Drag Image A

Face Image B (Reference)

Click or Drag Image B

Face Engine Pipeline Performance Benchmark

Evaluates SCRFD detection and ArcFace 512-D embedding extraction throughput on your hardware.

Biometric Accuracy & Error Rate Trade-off Analysis

False Acceptance Rate (FAR) vs. False Rejection Rate (FRR) curves across cosine similarity thresholds for ArcFace.

Threshold (\(\theta\)) False Acceptance Rate (FAR) False Rejection Rate (FRR) True Acceptance Rate (TAR) Operational Assessment

📐 Biometric Performance Principles

FAR (False Acceptance Rate)

Probability that an unauthorized/unknown face is falsely accepted as a registered user. Decreases as threshold increases.

FRR (False Rejection Rate)

Probability that a legitimate registered user is incorrectly rejected. Increases as threshold increases.

Recommended Threshold (0.50 - 0.60)

Provides optimal balance (FAR < 0.001%, TAR > 99.5%) for enterprise access control and Logitech webcam conditions.

Recognition Audit Trail (SQL Server)

Log ID Timestamp User ID Name Similarity Status Latency