Research On Multimodel Biometrics Based On Fingerprint And Finger Vein

In modern society, with the rapid development of information and networktechnology, the information security becomes more and more important. Due toconveniences, safety and reliability, biometric authentication based on the physical orbehavioral attributes of an individual is gradually replacing traditional identificationmethods, such as key, password, ID card and etc. However, uni-biometric systems haveto contend with a variety of problems such as noisy data, intra-class variations,restricted degrees of freedom, non-universality, spoof attacks, and unacceptable errorrates. Some of these limitations can be handled by deploying multi-biometric systemsthat integrate the evidence presented by multiple sources of information. Fingerprint isthe most widely used biometric trait and vein has high security. Thus, it is of greatpractical value and academic significance to research the methods on multimodelbiometrics based on fingerprint and finger vein.This paper starts from the fingerprint and finger vein identification subsystems,carries out some elementary research on image processing, feature extraction andmatching algorithms. Then, the fusion recognition algorithms are designed and in turnthe fusion identification System is built. The main innovations of this paper are asfollows.①A fusion algorithm based on dynamic weighting at feature extraction level isproposed. Neighborhood elimination and reservation of points belonging to specificregions are implemented to handle the problem of curse of dimension. According to theresults of features evaluation, dynamic weighting strategy is introduced for the fusion ofbiometrics. The weight of excellent features in fusion is improved, aiming to weakenthe influence of low quality and false features so that better effects of fusion can beachieved. Experimental results show that our scheme is remarkable advantage inperformance over uni-biometric systems. The proposed method of quality evaluation iseffective to classify for the specific features.②A fusion algorithm based on Classifier-Sequence (CS-Fusion) at matching levelis proposed. The minutiae and concatenated features of double-model traits are togetheras the classification criterions. Experimental results show that the algorithm achieves1.61%EER (Equal Error Rate), improving verification performance and securitysignificantly. Furthermore, the CS-Fusion algorithm both improves the performance greatly and44.83%identification speed than direct Sum-based fusion algorithm.③Multimodel biometric system based on fingerprint and finger Vein is built uppreliminarily. Using the toolbox of GUIDE to design and implement the multimodelbiometric system based on the CS-Fusion algorithm, further to verify the effectivenessof the proposed algorithms.