Research On Dorsal Hand Vein Recognition Based On Convolutional Neural Network

Today,information technology is developing rapidly,people’s lives are more and more convenient.While information technology brings us convenience,it also brings great challenges to our information security.Therefore,modern society urgently needs a more safe,efficient and accurate biometric authentication method.Compared with fingerprint,iris,face recognition and other biometric authentication methods,vein recognition is more safe and effective.Vein recognition is divided into finger vein,palmar vein and dorsal hand vein.The dorsal hand vein has more abundant features,and the vein is located in the superficial layer,so the features are easier to extract.Therefore,dorsal hand vein recognition has great research significance.In this thesis,aiming at the problem of insufficient data of hand dorsal vein,a device was built independently to collect images.At the same time,image enhancement and feature extraction of hand dorsal vein were studied.Finally,the veins were classified with convolutional neural network.The main innovation points of this thesis are as follows:(1)In view of the problems that most of the existing open venous datasets refer to venous datasets,and the open venous datasets of the back of the hand are few and of low quality,and are not easy to be flexibly applied to their own experiments due to their size,resolution and other limitations in the experiment,this thesis,based on the near-infrared imaging principle,uses the reflection method to build a device that can collect the image of the back of the hand vein.The device is simple to operate.By freely adjusting the light intensity through the switch,the lift table and the camera focus can be adjusted together to collect high-quality hand vein,thus providing data guarantee for subsequent experimental research.(2)In view of the problem that the median filter in the process of removing the noise in the hand vein image enhancement causes the image to become blurred,thus losing the hand vein features,the median filter algorithm is improved,and the conditions for pixel removal are restricted,so that the image can not only filter out the noise,but also retain the details of the vein to the greatest extent,and improve the accuracy of subsequent recognition.(3)In view of the problem that the median filter in the process of removing the noise in the hand vein image enhancement causes the image to become blurred,thus losing the hand vein features,the median filter algorithm is improved,and the conditions for pixel removal are restricted,so that the image can not only filter out the noise,but also retain the details of the vein to the greatest extent,and improve the accuracy of subsequent recognition.(4)In order to solve the problem that the training time of convolution neural network is long when the data size is large,this thesis has done size normalization for the thinned vein,which has smaller vein input size.At the same time,several convolution neural networks have been improved to enhance their ability to extract features from small size images.