An End-to-End Optimized Lensless System for Privacy-Preserving Face Verification.

Area of Science:

• Computer Vision
• Optics
• Biometrics

Background:

• Lensless cameras offer compact, cost-effective imaging by encoding light directly onto sensors.
• Traditional lensless face verification involves reconstruction, posing privacy risks and high computational costs.
• Existing methods struggle with lensless camera specific challenges like face detection and alignment.

Purpose of the Study:

• To develop an end-to-end privacy-preserving face verification system for lensless camera captures.
• To eliminate the need for intermediate face reconstruction, thereby enhancing user privacy.
• To improve the efficiency and performance of lensless face verification systems.

Main Methods:

• Proposed an end-to-end optimization approach for direct verification on encoded lensless data.
• Introduced a novel face center alignment scheme tailored for lensless imaging.
• Implemented an augmentation curriculum and knowledge distillation for robustness and performance.
• Ensured the entire software pipeline operates on encoded data, preventing visible face generation.

Main Results:

• The end-to-end method significantly outperforms traditional two-stage lensless verification.
• Demonstrated enhanced privacy by eliminating intermediate face reconstruction.
• Achieved improved computational efficiency compared to existing lensless verification techniques.
• Validated performance through evaluations in both simulated and real-world environments.

Conclusions:

• End-to-end optimization is a viable and superior approach for lensless face verification.
• The proposed techniques effectively address challenges unique to lensless imaging systems.
• This method enhances privacy and efficiency, making lensless cameras more suitable for sensitive applications.