Bynn Face Occlusion Detection

The Bynn Face Occlusion Detection model determines whether a person's face is occluded or obscured in an image. This model is essential for identity verification workflows where clear face visibility is required for accurate identification.

The Challenge

Identity verification fails silently when faces are obscured. Users submit photos wearing sunglasses, masks, or hats. Images are blurry, poorly lit, or awkwardly cropped. Facial recognition systems return low confidence scores or false matches, but cannot explain why—leading to frustrated users and support escalations.

Early detection of face occlusion prevents wasted processing and poor user experience. Rather than running expensive facial recognition on unsuitable images, platforms can immediately prompt users to resubmit with clear face visibility. This quality gate reduces verification failures, speeds processing, and improves completion rates for onboarding flows.

Beyond identity verification, face occlusion detection enables critical security applications. CCTV systems can flag individuals deliberately concealing their identity—ski masks, balaclavas, or face coverings inappropriate for the environment. Early detection of masked individuals in banks, schools, or retail environments can trigger alerts before incidents escalate, potentially preventing robberies or violent attacks.

Model Overview

When provided with an image containing a person, the detector analyzes whether the face is clearly visible or obstructed by various factors including masks, sunglasses, hand positions, image quality issues, or framing problems.

Achieving 93.0% accuracy, the model uses Bynn's Visual Language Model technology to understand both physical obstructions and image quality factors that prevent clear face visibility.

How It Works

The model evaluates multiple factors that can obstruct face visibility:

• Physical obstructions: Masks, sunglasses, hands, and other objects covering the face
• Image quality: Blur, pixelation, and lighting issues affecting visibility
• Framing issues: Face cut off from frame or image rotation problems
• Digital filters: AR filters, face masks, or other virtual overlays

Response Structure

The API returns a structured JSON response containing:

• occluded: Boolean value (true/false) indicating whether the face is occluded

Detection Categories

Occluded (true)

Face is considered occluded when:

• Blurry image where person's face is not clearly visible
• Pixelated image where person's face is not clearly visible
• Image is too dark to clearly see the person's face
• Face is partially or fully occluded by another body part or person's pose
• Part or all of the face is cut off from the frame
• Face is covered by masquerade mask or other face masks
• Face is occluded with animal face filter or other virtual filters
• Face is partially covered by sunglasses or tinted glasses
• Image appears to be rotated or not right side up

Not Occluded (false)

Face is considered not occluded when:

• Person's full face is clearly visible without any rotation to the image
• Person's full face is clearly visible without blurriness or pixelation
• Person's full face is clearly visible and well lit
• Person's full face is clearly visible without other body parts covering the face
• Person's full face is clearly visible without being cut off from the frame
• Person's full face is clearly visible without any filters on top
• Person's full face is clearly visible without any face masks
• Person's full face is clearly visible without sunglasses or tinted glasses
• Person's full face with transparent/clear glasses is clearly visible

Performance Metrics

Use Cases

• KYC Verification: Ensure identity document selfies have clear, unobstructed face views
• Profile Photo Validation: Verify user profile photos meet visibility requirements
• Biometric Enrollment: Pre-check images before facial recognition enrollment
• Security Screening: Detect attempts to hide identity with masks or obstructions
• Photo Quality Control: Filter out low-quality or improperly framed photos
• Age Verification: Ensure face is visible for age estimation workflows

Known Limitations

Important Considerations:

• Transparent Glasses: Clear prescription glasses are not considered occlusion; tinted lenses are
• Partial Occlusion: Minor obstructions may not always be flagged if face is mostly visible
• Multiple Faces: Analysis focuses on the primary/largest face in the image
• Cultural Considerations: Some head coverings may be flagged depending on face visibility
• Artistic Photos: Intentional artistic lighting or poses may be flagged as occlusion

Disclaimers

This model provides face visibility assessment, not identity verification.

• Pre-screening Tool: Use as a quality gate before identity verification, not as verification itself
• Resubmission Flow: Flagged images should prompt users to submit clearer photos
• Context Awareness: Some occlusions (medical masks, religious coverings) may require policy-specific handling
• Accessibility: Consider alternative verification methods for users who cannot remove certain face coverings
• Combined Workflows: Use alongside face detection and liveness checks for comprehensive verification

Best Practice: Integrate occlusion detection early in verification workflows to provide immediate feedback and improve submission quality.