Research On Finger Vein Recognition

As human beings enter the information age,more and more attention has been paid to information security,so it is of great significance to study the safe and efficient biometric technology.Finger vein recognition has become a research hotspot in the field of biometric recognition because of its internal characteristics,living recognition and high security.This paper studies finger vein recognition based on traditional methods and deep learning,and discusses the entire process of image preprocessing,feature extraction and recognition matching in detail.The main contents of this paper are as follows:(1)The characteristics of common biometric recognition technologies are summarized,and the advantages of finger vein recognition are analyzed.This paper also expounds the research status of finger vein image acquisition,image preprocessing and feature extraction,and then discusses the research difficulties and overall process framework of finger vein recognition.(2)The preprocessing algorithm of finger vein image is studied.Aiming at the problems of finger rotation and offset during image acquisition,a rotation correction algorithm is proposed based on finger midline,which captures the two midpoints of the edge position to fit the midline to calculate the angle deviation and correct the image.At the same time,the finger joints are located based on the difference in image brightness,which are used as baselines to intercept the image to effectively deal with the finger offset.(3)A series of feature extraction and matching algorithms based on local binary pattern(LBP)are studied.The traditional LBP Operators ignore the spatial position information when extracting the image features,so a blocking strategy is introduced which divides the image into several sub-images,and cascades the LBP features of each sub-image to obtain the blocked LBP features of the whole image,so as to improve recognition accuracy.However,the increase in the number of blocks leads to rapid growth of data dimension,so it is further proposed to combine the blocked LBP with principal component analysis(PCA).After extracting the features of blocked LBP,PCA dimension reduction is carried out and then recognition matching is practiced so that the recognition speed could be improved effectively.At the same time,this paper attempts to take the operation for a single image instead of centralized processing of the entire database,in order to meet the actual needs of industrial applications better.(4)The application of convolutional neural network(CNN)in finger vein recognition is studied.In view of the shortage of database samples,reliable data enhancement strategies are proposed,such as translation,rotation,noise addition and so on.At the same time,a simple CNN model with small convolution cores and large steps is designed based on classical networks,which can reduce the parameter scale and speed up the calculation process.Then,the LBP and CNN are combined to train the network by using the LBP characteristic map of finger vein image.Compared with the pixel level knowledge of original image,the LBP feature,which represents image texture,can be used as the training input to make the network obtain better learning effect.(5)Experiments are conducted on two public finger vein databases and a self-built database.Experiments show that the LBP operators perform very well in the field of finger vein recognition;the blocking strategy can greatly improve the recognition accuracy;increasing the number of training samples can make up for the lack of image quality,and effectively improve the recognition rate when the quality of samples is poor.The series of algorithms based on LBP can achieve the highest recognition accuracy of 100%,94% and 100% on three databases.At the same time,the combination of blocked LBP and PCA dimension reduction is a very potential attempt.The introduction of PCA has doubled the recognition speed.Although it is still necessary to gain improvement in recognition rate,it undoubtedly provides a meaningful practical idea for the application of industry.In addition,convolutional neural network has achieved excellent results in all three databases,which shows that CNN can break the bottleneck of image quality,and has great advantages over traditional LBP algorithms.Moreover,the method of combining LBP with CNN shows better recognition performance compared with ordinary CNN model,which can improve convergence rate and recognition accuracy.