Bynn Document Liveness Detector

The Bynn Document Liveness Detector determines whether a document or portrait image was captured directly from the physical document (live) or photographed from a screen display. This model is critical for KYC/KYB workflows where document authenticity verification is essential.

The Challenge

Identity fraud has evolved beyond simple document forgery. Fraudsters now photograph legitimate documents displayed on screens—using images found online, stolen from data breaches, or obtained through social engineering. This "screen capture" attack bypasses traditional document verification because the underlying document may be genuine, even though the submission is fraudulent.

The technique is devastatingly simple: find or steal someone's ID photo, display it on a phone or monitor, and photograph it for submission. Without liveness detection, verification systems cannot distinguish this attack from legitimate document capture. Financial institutions, cryptocurrency exchanges, and any platform requiring identity verification face millions in fraud losses from this single attack vector.

Model Overview

When provided with an image of a document, the detector analyzes visual characteristics to identify signs of screen capture versus live document photography. The model looks for specific indicators like screen bezels, reflections, pixel patterns, and unnatural lighting that reveal when a document has been photographed from a digital display.

Achieving 93.0% accuracy, the model uses Bynn's Visual Language Model technology trained on document forensics to detect subtle visual artifacts that indicate screen capture fraud.

How It Works

The model analyzes multiple visual indicators to determine document liveness:

Screen Capture Indicators

• Screen bezels, frame edges, or device borders visible
• Screen reflections or glare
• Visible curvature from screen surface
• Another device (phone, monitor) clearly visible in the image
• Unnatural lighting typical of screen displays

Live Document Indicators

• Natural paper or card texture
• Natural lighting and shadows
• Document appears to be the original physical item

Response Structure

The API returns a structured JSON response containing:

• liveness: Classification - "live", "computer_screen", "phone_screen", or "other_screen"
• confidence: Confidence level - "low", "medium", or "high"
• indicators: Array of detected visual signs that led to the classification

Classification Levels

Performance Metrics

Use Cases

• KYC Verification: Detect fraudulent identity document submissions where documents are displayed on screens
• KYB Compliance: Verify business documents are authentic physical copies, not digital reproductions
• Onboarding Security: Add liveness checks to customer onboarding workflows
• Fraud Prevention: Block common identity fraud techniques involving screen-displayed documents
• Remote Verification: Support secure remote document verification without physical presence
• Financial Services: Meet regulatory requirements for document authenticity in banking and fintech

Known Limitations

Important Considerations:

• High-Quality Displays: Very high-resolution displays with careful lighting may be harder to detect
• Cropped Images: If screen edges are cropped out of the image, detection may be more challenging
• Printed Copies: The model detects screen captures, not printed copies of digital documents
• Image Quality: Very low resolution or heavily compressed images may reduce detection accuracy
• Conservative Design: The model is tuned to avoid false positives; it prefers classifying as "live" when uncertain

Disclaimers

This model provides probability-based classifications, not definitive fraud determinations.

• Layered Security: Use as one layer in a multi-factor fraud detection strategy
• False Positive Prevention: The model is designed to minimize false rejections of legitimate documents
• Human Review: Screen capture detections should be reviewed, especially for high-value transactions
• Indicator Review: Check the indicators array to understand why a classification was made
• Complementary Checks: Combine with other document verification methods (OCR, tampering detection, etc.)

Best Practice: Use liveness detection as part of a comprehensive document verification workflow that includes multiple fraud detection signals.