摘要
鉴于传统局部二进制模式(local binary pattern, LBP)算法对光照方向的变化非常敏感的问题,本文提出一种融合旋转不变模式的 LBP算子与 B2DPCA技术的手指静脉识别方法。首先提取手指静脉图像子块的LBP纹理谱特征,然后采用双向二维主成分分析方法对 LBP特征向量构成的特征矩阵进行有效的降维处理,再通过比对降维后的待识别静脉图像特征向量与其他样本的特征向量之间的欧式距离来实现最终的样本分类。通过在天津市智能实验室静脉库及马来西亚理科大学 FV-USM静脉库上进行实验验证,在不同训练样本数量下比较了 8种算法的识别性能,相比于单一的 LBP特征提取算法、经典降维算法和 LBP与经典降维组合特征提取算法,该方法的识别率有很大的提高,证明了本文方法的有效性。
By considering the sensitivity of the traditional local binary pattern (LBP) algorithms while varying the illu-mination,this study proposes a finger vein recognition method using a rotation invariant LBP operator and B2DPCA.This method initially extracts the LBP texture spectrum feature of the image block of a finger vein,uses a bidirectional two-dimensional main component analysis method to effectively reduce the dimension of the eigenmatrix comprising the LBP eigenvectors,and finally classifies the final samples by comparing the Euclidean distance between the vein im-age eigenvectors that are to be identified and the eigenvectors of other samples after dimension reduction.The experi-ments were implemented on the finger vein image databases obtained from the Tianjin Intelligence Laboratory and from the FV-USM database of the University of Science,Malaysia.Further,eight methods with different numbers of training samples are compared,which exhibit that the fusion features that are proposed by this study perform considerably better than the single LBP operator,single traditional dimension-reduced methods,and the fusion of LBP and traditional di-mension-reduced algorithms.Additionally,the recognition rate of the generated method was observed to significantly improve.This indicated that the analysis method proposed in this study is proper and effective.
作者
胡娜
马慧
湛涛
HU Na;MA Hui;ZHAN Tao(College of Electronic Engineering,Heilongjiang University,Harbin 150001,China)
出处
《智能系统学报》
CSCD
北大核心
2019年第3期533-540,共8页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(61573132)
黑龙江省高校基本科研业务费项目(HDRCCX-201602)
黑龙江省高校重点实验室开放基金项目(DZGC201610)
关键词
手指静脉识别
特征提取
LBP纹理特征
二维主成分分析
双向二维主成分分析
欧氏距离
图像特征向量
降维
finger vein recognition
feature extraction
local binary patterns
two-dimensional principal component
bid-irectional two-dimensional principal component analysis
euclidean distance
image feature vector
dimensionality re-duction
