基于多模式LBP与深度森林的指静脉识别  

Finger Vein Recognition Based on Multi-mode LBP and Deep Forest

作　　者：刘广东 邱晓晖[1] LIU Guang-dong;QIU Xiao-hui(School of Communications and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《计算机技术与发展》2018年第7期83-87,共5页Computer Technology and Development

基　　金：江苏省自然科学基金(BK2011789);东南大学毫米波国家重点实验室开放课题(K201318)

摘　　要：深度森林(gc Forest)是基于深度模型提出的级联随机森林集合方法,以解决深度学习网络模型中对大样本训练数据和对设备要求过高的问题。深度森林不像深度神经网络那样具有很多的调节参数,对训练模型的选取需要耗费大量的时间与精力,gc Forest允许使用者可以根据设备的资源决定训练的耗费,且能自适应地调节训练模型层数。指静脉图像含有丰富的纹理信息,文中基于多模式LBP提取指静脉图像的基本LBP特征,统一模式LBP分块直方图特征并将它们与深度森林结合取得的识别率达到99.46%,训练时间大幅减少,并解决了gc Forest在旋转适应性方面的不足。与随机森林分类器(random forest)、KNN分类器、支持向量机分类器(SVM)、罗格斯特回归分类器(logistic regression)等进行比较,证明了深度森林识别器的有效性。The deep forest,named as gc Forest,is a cascade random forest ensemble method based on the deep learning model,in order to solve the problem that the deep learning network model usually requires large scale of training samples and the threshold of the deep learning is too high for personal research.Gc Forset doesn＇t have lots of tuning parameters like convolution neural networks which cause the selection of the training models taking lots of time and effort.Gc Froest allows users to determine the cost of training based on the resources of the device,and can adjust the number of training model layers in an adaptive manner.The finger-vein image contains rich texture information.Based on multi-mode LBP,we extract the basic LBP features and unified model LBP partitioned histogram features in the finger-vein image and combine them with the depth of the forest to reach the recognition rate by 99.46%,sharply reducing training time and solving the shortcoming on the adaptability for gc Forest.Compared with random forest,KNN classifier,SVMand logistic regression,the effectiveness of gc Forest recognizer is proved.

关 键 词：指静脉识别 深度森林 LBP 特征提取 随机森林 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]