Fingerprint Presentation Attack Detection Utilizing Spatio-Temporal Features.

Area of Science:

• Biometrics and Security
• Pattern Recognition
• Computer Vision

Background:

• Presentation attacks pose a significant threat to fingerprint recognition systems.
• Existing methods struggle to effectively detect diverse presentation attack species.
• Dynamic analysis of fingerprint patterns is crucial for robust security.

Purpose of the Study:

• To develop and evaluate a novel mechanism for dynamic presentation attack detection in fingerprint systems.
• To enhance the security of fingerprint recognition against various spoofing techniques.
• To validate the proposed method using standardized experimental protocols.

Main Methods:

• Utilized five spatio-temporal feature extractors to capture dynamic fingerprint pattern variations.
• Integrated ridge/valley patterns with temporal dynamics from fingerprint videos.
• Employed a Support Vector Machine (SVM) classifier with a second-degree polynomial kernel for classification.
• Conducted experiments and evaluations adhering to the ISO/IEC 30107-3:2017 standard.

Main Results:

• The proposed approach demonstrated high efficiency in detecting presentation attacks.
• Achieved a low BPCER (Best Possible Classification Error Rate) of 1.11% for optical sensors.
• Achieved a BPCER of 3.89% for thermal sensors, both at a 5% APCER (Actual Presentation Attack Detection Error Rate).

Conclusions:

• The novel spatio-temporal feature extraction and SVM classification effectively detect fingerprint presentation attacks.
• The method offers robust performance across different sensor types (optical and thermal).
• The system meets high security standards, significantly reducing the likelihood of successful spoofing attempts.