fPADnet: Small and Efficient Convolutional Neural Network for Presentation Attack Detection.

Area of Science:

• Biometrics and Security
• Computer Vision
• Machine Learning

Background:

• Fingerprint authentication is widespread, necessitating robust presentation attack detection (PAD) to counter fake fingerprints.
• Existing PAD methods often struggle with a balance between accuracy and computational demands.
• Developing efficient and accurate fake fingerprint detection is critical for secure biometric systems.

Purpose of the Study:

• To propose an efficient and accurate Convolutional Neural Network (CNN) model for fingerprint presentation attack detection (PAD).
• To introduce a novel network architecture, fPADnet, integrating Fire and Gram-K modules for enhanced feature extraction.
• To address the trade-off between accuracy and computational complexity in existing PAD algorithms.

Main Methods:

• Developed fPADnet, a CNN-based network incorporating SqueezeNet-inspired Fire modules and Gram-K modules for texture analysis.
• Gram-K modules were utilized to extract discriminative texture features essential for differentiating real from fake fingerprints.
• The combined architecture aims for a compact and computationally efficient model.

Main Results:

• fPADnet achieved an average detection error rate of 2.61% on public LivDet databases (2011, 2013, 2015).
• The proposed method demonstrates accuracy comparable to state-of-the-art techniques.
• Significant reductions in network size and processing time were observed compared to existing methods.

Conclusions:

• fPADnet offers a highly effective solution for fingerprint presentation attack detection.
• The integration of Fire and Gram-K modules provides a compact yet powerful network architecture.
• The method successfully balances high accuracy with reduced computational complexity, making it suitable for real-world applications.