Implementation Of Finger Vein Recognition System Based On Extended Convolutional Neural Network And Metric Learning

As a new biometric technology,finger vein recognition has attracted more and more attention due to its unique recognition advantages,such as high security,non-contact,and difficult to steal.But finger vein recognition algorithms based on traditional image processing are usually more sensitive to image quality and finger pose changes,making it difficult to extract robust vein features.The vein feature information extracted by the finger vein recognition algorithm based on deep learning is not enough to represent vein features with complex texture structures,and the Commonly used distance measurement algorithms mainly uses the distance or directional angle between features as the similarity of the features.Such a measurement algorithm cannot measure the feature similarity from the structural essence.In order to solve the problem of insufficient vein feature information extracted by finger vein recognition algorithm based on deep learning.This paper proposes a finger vein recognition algorithm based on extended convolutional neural network and metric learning.In addition,in order to make up for the problem that commonly used distance measurement algorithms cannoteffectively measure the similarity between vein features,a Wasserstein distance measurement algorithm is used.The main research contents of this article are summarized as follows:(1)By analyzing the problem of insufficient vein feature information extraction from the traditional Convolutional Neural Networks(ConvNet)network,this paper first builds a learning network to extract finger vein features.This network expands the convolutional neural network width and depth to improve the network's ability to extract vein features.Secondly,a reasonable loss function is designed according to the constructed learning network.This article uses the TriHard metric learning function as the network loss function.Third,during network training,the overall loss of the network is calculated through forward propagation,and the stochastic gradient descent(SGD)back propagation was used to update the parameters of the ConvNet network.(2)By analyzing the problem that commonly used distance measurement algorithms can not effectively measure the similarity between vein features,this paper uses Wasserstein distance measurement algorithm to calculate the similarity between two feature vectors.This algorithm can not only improve the similarity between homologous veins,but also reduce the similarity between heterologous veins.(3)In order to further research and develop the vein recognition algorithm.In this paper,a finger vein recognition system is designed and implemented.This system provides a finger vein visualization interface and can save the vein features extracted using the algorithm in this paper to a local database.At the same time,it can provide the recognition result information during the vein recognition phase.This paper uses the public vein datasets FV-USM and MMCBNU 6000 for testing.The experimental results show that the extended convolutional neural network and metric learning algorithm proposed in this paper have a higher accuracy rate in finger vein recognition algorithms.At the same time,the accuracy of vein recognition using Wasserstein distance measurement algorithm is higher than that of other commonly used distance measurement algorithms.