An efficient multi-biometric cancellable biometric scheme based on deep fusion and deep dream

Area of Science:

• Biometrics and Security Engineering
• Artificial Intelligence in Cybersecurity
• Pattern Recognition and Machine Learning

Background:

• Biometric systems are increasingly targeted by sophisticated criminal attacks.
• Existing biometric security methods require robust solutions to protect sensitive personal data.
• The need for secure, tamper-proof biometric templates is paramount for reliable authentication.

Purpose of the Study:

• To develop and validate a novel multi-biometric cancellable scheme (MBCS).
• To enhance the security and integrity of biometric data against unauthorized access and manipulation.
• To create an aggregate, tamper-proof cancellable template using deep learning.

Main Methods:

• A multi-biometric approach fusing fingerprint, finger vein, and iris data.
• Utilizing a deep learning model (Inspection V3) for template generation.
• Implementing a multi-exposure technique for enhanced feature extraction.
• Conducting extensive quantitative and qualitative evaluations.

Main Results:

• Achieved high performance metrics, including 99.158% NPCR and 0.909 UIQ.
• Demonstrated favorable comparison against state-of-the-art methods.
• Generated tamper-proof cancellable templates with strong security properties.
• Recorded low error rates (e.g., 0.079 SSIM) and high image quality (e.g., 24.523 dB PSNR).

Conclusions:

• The proposed MBCS offers a robust and secure solution for biometric authentication.
• The scheme effectively protects biometric modalities against various security threats.
• Future work aims to refine the MBCS for real-time applications in critical sectors like airports and banks.