Lensless facial recognition with encrypted optics and a neural network computation

Area of Science:

• Computer Vision
• Biometric Authentication
• Deep Learning

Background:

• Face recognition is crucial for biometric authentication.
• Conventional lens-based imaging raises privacy concerns due to spatial fidelity.
• Existing methods may require lens aberration correction and reveal facial features.

Purpose of the Study:

• To develop a privacy-preserving face recognition system.
• To utilize coded masks for image encryption and deep learning for identification.
• To overcome privacy issues associated with traditional face recognition.

Main Methods:

• Employed a coded mask for image encryption based on point spread function engineering.
• Utilized deep neural network computation for feature extraction and identification.
• Generated a dataset with practical photographing and data augmentation.

Main Results:

• Achieved 100% recognition accuracy on real-time measurements.
• The system demonstrated robustness to a wide depth of field (60-cm hyperfocal distance).
• The framework adapted to significant pose variations (0 to 45 degrees).

Conclusions:

• The proposed data-driven approach ensures privacy by creating non-interpretable image representations.
• No physics priors or lens aberration correction are needed for successful identification.
• This method offers a secure and accurate solution for biometric face recognition.