Effect of Blurred Fingerprints in Biometric Identification using Machine Learning

Abstract

Biometrics has transformed identification using human traits, but challenges like privacy concerns and security vulnerabilities persist. Fingerprint biometrics, crucial for uniqueness, has evolved with digitalization and machine learning. The paper investigates blurring effects on fingerprint features, proposing machine learning for comparative minutiae-based matching. Gaussian blur impact on identification accuracy is studied, with a decline observed beyond a standard deviation (SD) of 0.3. FLANN matching score remains 100 % for SD in the range 0.1-0.3. The diminishing matching ratio and modified minutiae spatial pattern with increasing SD highlight the influence of blurring. The study assesses a machine learning system's tolerance to blurring, revealing poor matching beyond an SD of 0.4. Generally, the blurring is introduced because of skin scattering, optical blur, or motion blur, emphasizing the need for a pilot mechanism with standard reference fingerprints before scanning for developing future-ready fingerprint scanners.