Research On Finger Vein Feature Extraction And Matching Algorithm Based On Differential Geometry

As a highly secure and stable biometric identification technology,vein recognition has gained extensive recognition.However,there are still challenges in feature extraction and matching efficiency during the process.This paper focuses on studying the feature extraction algorithms of veins and matching efficiency of recognition systems from these two aspects.The work includes:(1)proposing two feature extraction algorithms based on differential geometry to solve the problem of poor feature extraction caused by low-quality images.The first method involves calculating the discrete curvature values of vein contour images based on planar differential geometry.The second algorithm uses the properties of three-dimensional differential geometry to extract vein features from the concave region of the transformed vein image.The two algorithms are compared with Gabor filtering algorithms,repeated line tracking algorithms,and maximum curvature algorithms in terms of performance using the FV-USM and MMCBNU＿6000 datasets.Results show that the proposed algorithms outperform the other three algorithms in multiple performance measures,with an average error recognition rate,error rejection rate,and equal error rate of 3.05%,3.17%,and2.80%,respectively,which is a 6.30% performance improvement compared to the best-performing algorithm.(2)As for matching efficiency of vein recognition systems,the paper introduces a feature matching algorithm based on Earth Mover’s Distance(EMD).This algorithm uses discrete curvature as a distance feature and calculates the similarity of specific areas using EMD algorithm to compare the curvature distributions of vein contour points.In order to improve the efficiency of the algorithm,the paper also refers to a fast EMD algorithm to implement high-efficiency matching.The EMD algorithm outperforms previous feature point matching and vein pattern matching algorithms in terms of matching precision on the FV-USM and MMCBNU＿6000 datasets with a 10.58% performance improvement compared to the best-performing algorithm,and the improved fast EMD algorithm has a 4.61%higher matching accuracy than the EMD algorithm.