基于灰度不均匀矫正和SIFT的手指静脉识别方法  

Finger vein recognition based on intensity inhomogeneity correction and scale invariant feature transform

作　　者：孟宪静 袭肖明 杨璐 尹义龙[2] 

机构地区：[1]山东财经大学计算机科学与技术学院,济南250014 [2]山东大学计算机科学与技术学院,济南250101

出　　处：《南京大学学报（自然科学版）》2018年第1期1-10,共10页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(61701280;61573219;61671274);山东省自然科学基金青年项目(ZR2016FQ18);山东省高等学校优势学科人才团队培育计划

摘　　要：基于手指静脉的身份识别以其方便性和安全性奠定了其在生物特征识别中的优势地位.在手指静脉识别方法中,尺度不变性特征(Scale Invariant Feature Transform,SIFT)虽然普遍被认为效果不佳,但鉴于SIFT在自然图像中取得的良好效果,在分析了手指静脉图像的质量和结构特点之后,设计了一种基于灰度不均匀矫正和SIFT的手指静脉识别方法.首先,根据手指静脉图像对比度低、模糊等特点,利用灰度不均匀矫正增强图像细节;其次,考虑到在尺度不变性特征的匹配过程中,相似特征点的存在也会影响手指静脉识别的性能;第三,在匹配的过程中,还考虑了潜在的特征匹配点.基于灰度不均匀矫正和潜在特征匹配点的手指静脉识别方法取得了良好的识别效果,在香港理工大学手指静脉库(PolyU Finger Vein Database)上六折交叉验证的等错误率(Equal Error Rate,EER)从0.0358降低到了0.0006,表明了方法的有效性.Finger vein,one of the most promising biometric pattern,has received considerable attention from researchers because of its convenience and security.Scale Invariant Feature Transform（SIFT）is an algorithm to detect local features in images,which is rotation and scale invariant,and is widely applied in detection and recognition tasks.However,the results of finger vein recognition methods based on SIFT are demonstrated to be unsatisfactory.To deal with this problem,we propose a finger vein recognition method based on SIFT with intensity inhomogeneity correction and a new matching technique.The finger vein image is firstly processed by intensity inhomogeneity correction based upon the observation that the obscured image quality is bad for SIFT key-points detection.The key-points extracted are usually unstable and small in amount.After the correction of intensity inhomogeneity,the textures in images can be enhanced,and key-points can be correctly detected and be more stable.In the SIFT matching process,the existing of too many similar key-points can also degrade the performance.This is because possessing which is more than one similar key-points may lead to false matching.We believe key-points match successfully with more than one key-point should also be considered in matching,and name it as potential matching key-point.Accordingly,we calculate the matrix of matching scores for two images,on which the distribution of potential matching key-point is assessed.Finally,the matching scores is calculated as normalized number of potential matching key-points.Extensive experiments have been conducted on the PolyU and MLA databases.The Equal Error Rate（EER）under six-fold cross validation on the PolyU database is reduced to 0.0006 from 0.0358,which indicates the effectiveness of the proposed method.

关 键 词：手指静脉识别 尺度不变性特征 潜在特征匹配点 灰度不均匀矫正 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]