A Robust Fingerprint Identification Method by Deep Learning with Gabor Filter Multidimensional Feature Expansion

摘要

Traditional fingerprint methods based on minutiae matching perform well for the acquisition of large area fingerprint. But the accuracy rate and the robustness of small area fingerprint decreases obviously when contains less minutia. Aiming at solving the above problem, a small area fingerprint matching method based on Convolution Neural Network (CNN) which selecting the center block of fingerprint as the region of interest (ROI) after preprocessing and using the Gabor filter to extract feature as multidimensional feature extension named ROIFE_CNN (ROI of fingerprint feature extension recognition of CNN) is proposed to enhance robustness. Experiments show that the accuracy of small area fingerprint classification based on CNN is enhanced.