Study On Feature Extraction Method Of Finger Vein Image

As personal information security issues have become more and more important,biometric-based identification technologies such as fingerprint recognition,face recognition,and iris recognition have received more and more attention.In recent years,finger vein recognition which is a new type of biometrics has become a research hotspot in the field because of its unique advantages.Compared with traditional biometric technology,the finger vein recognition technology has the following advantages:(1)easier to use:simple,non-contact authentication of the collection device;(2)living body identification:only the veins of blood flow can be certified and recognized;(3)Anti-counterfeiting:venous blood vessels are hidden in the human body and difficult to be stolen and forged;(4)Accurate and reliable:the vein image is unique and stable,free from diseases,climatic conditions and finger surface conditions.Therefore,finger vein recognition technology has gradually become a hot topic in the field of biometrics.The finger vein recognition technique consists of four steps:(1)acquisition of the finger vein image;(2)preprocessing of the image;(3)feature extraction;and(4)matching and recognition.Among them,the quality of feature extraction will have a great impact on the recognition effect.Therefore,the research in the field of finger vein recognition is mostly focused on the method of feature extraction.This thesis mainly studies the method of finger vein image feature extraction,and proposes a feature extraction method based on local descriptors which is more suitable for finger vein recognition.In addition,the deep learning method is applied to the feature extraction of finger vein images.The main work of the thesis is summarized as follows:(1)Finger vein recognition based on the double orientation line feature weber local descriptor.Combined with the characteristics of the finger vein image,the weber local descriptor is improved,and a Double orientation Line feature Weber Local Descriptor(DLWLD)is proposed to improve the performance of finger vein image feature extraction.Firstly,when calculating the gray level difference of local regions,consider the influence of different neighborhood pixel points on the central pixel point,and construct a new differential excitation map.Then,the Modified Finite Radon Transform(MFRAT)is used to extract double orientation line feature.Finally,the improved differential excitation feature is combined with the line features of double orientations to obtain two feature vectors,and then the cross matching algorithm is used for identification to better evaluate the similarity between samples.The proposed local descriptors are applied in finger vein image recognition,and comparative experiments are performed in three public finger vein databases,and cross matching is performed under Euclidean distance.Experimental results show that compared with other popular local descriptors and existing improved WLD method,proposed method has a better recognition effect on finger vein recognition.(2)Finger vein recognition based on convolutional neural network.A convolutional neural network structure for finger vein feature extraction is designed to realize the end-to-end learning method,which enables the network to automatically learn the characteristic representation of the finger vein from the training samples and increase the discriminability of the features.The method uses the Inception-resnet module to stack the main body network(called FingerveinNet)to learn multi-scale feature representation,and uses center-loss as the cost function to obtain better detail distinguishing ability.Benefit from the characteristics of convolutional neural networks,this method is robust to changes of image such as rotation and translation,and can well solve the problem of sample rotation and translation in finger vein recognition.The experimental results show that this method has obvious advantages compared with the traditional finger vein recognition methods.