Font Size: a A A

Research On Finger Vein Recognition

Posted on:2015-03-02Degree:MasterType:Thesis
Country:ChinaCandidate:R Y XiaoFull Text:PDF
GTID:2268330431454467Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the arrival of the globalization development and information age, the information security requirement of modem society is much higher.Biometric recognitionas a new authentication technology, due to it’shigh level of convenience and safety, has beenregarded to more and more people.Among various kinds of biometric identifiers, finger vein recognition is a newly emerging biometrics technology. There are some advantages of the finger vein recognition method over other biometrics methods. Such as live body identification,non-contactand so on.But, finger vein recognition also has limitations, such as, low quantity and poor feature information of the finger vein images. In order to improve the accuracy of finger vein recognition system, in this thesis, we proposed some solutions. The main research contents are as follows:A typical finger vein recognition systemincludes four stages:image capturing, image preprocessing, feature extraction and matching. The feature extraction stage plays an important rule in the procedure. Among various kinds of finger vein feature extraction methods, Binary based method is an important class of methods.Due to its high discriminative power and invariant against any monotonic transformation of the gray scale, lots of modfied version of binary pattern has been researched and widely applied in finger vein recognition.In this paper, the Center-Symmetric local binary pattern (CSLBP) operator is firstly used as a feature extraction method for finger vein recognition.In addition, we also analyze the limitations of the LBP and CSLBP method, then proposed a new feature extraction method, called Multi-scale Block Center-Symmetric local binary pattern (MB-CSLBP).In this method, the comparison is performed based on average values of a sub-region instead of individual pixel. The proposed method fusesmicrostructures and macrostructures of the finger vein image, can provide a better image representation.Current binary pattern based finger vein matching methods treat every bit of the binary codes extracted by the feature extraction methods as equally important,this is simply irresponsible.In an ideal world, the binary codes extracted from different samples of the same object should have the same value, means0or1. But, we find that, some of them are stable, i.e., the bit values are all lor0. We called these’best bits’. On the contrary, some other bits are unstable, where some code series take the value of1, and the others value0.We called these’unstable bits’. We claim that the bits with different stable degrees should have different contributions for the final matching result. The’best bits’have a big contribution, so they should be given a high weight. And the ’unstable bits’should be given different weight corresponding there stability.For that reason, in this paper we propose a new finger vein recognition method based on a personalized weight map (PWM). The proposed method can significantly improve the recognition performance of the finger vein system as well as robustness.During these four different stages, there is a close relationship between feature extraction stage and matching stage. Vein vessel network is a very important vein pattern for finger vein recognition. Based on this pattern, lots of matcing algorithms have been used in the matching stageto evaluate the similarity between two finger vein images. But these existing matching algorithms have limitations. In this paper, we propose a novel matching strategy region-based axis projection (RAP) for finger vein recognition. In this method, we treat the finger vein vessel network as a distribution graph which consists of0and1. Firstly, the vein images are divided into small regions. Secondly, we concatenate the projection of the binay vein distribution curves on the x-axis and y-axis of each region, and finally we calculate the projections of the whole binary vein iamge.The proposed method can avoid image translation and rotation to some extent and gain a better performance.
Keywords/Search Tags:Finger VeinRecognition, Local Binary Pattern, Multi-scale BlockCenter-Symmetric Local Binary Pattern (MB-CSLBP), PersonalizedWeight Map (PWM), Region-based Axis Projection (RAP)
PDF Full Text Request
Related items