Research On Finger Vein Recognition Based On Deep Learning And Discrete Hash

Biometric technology,as an emerging identity authentication technology,has gradually replaced the traditional secure identity authentication technology and is widely used in various security fields.Finger vein feature recognition technology is a very important part of biometric recognition technology.It is widely concerned by scholars due to its characteristics such as live detection,not easy to forge,and high security.The traditional finger vein recognition method is sensitive to changes in finger vein image acquisition and has poor robustness.The linear retrieval algorithm of high-dimensional real-valued features will greatly decrease the recognition speed as the number of recognition increases,and the storage of a large number of finger vein feature templates will cause greater pressure on the device.In order to solve these problems,this paper proposes a finger vein recognition method based on deep residual network and discrete hash.The main research contents are as follows:?1?Aiming at the problems that the general neural network is not enough to extract the features of the finger vein image and the deep neural network is prone to network degradation,an improved deep residual network is proposed to extract the features of the finger vein image.The improved deep residual network uses a pre-activated residual module to avoid information loss during network transmission.The PRelu activation function is used instead of the original Relu function to minimize the loss of non-sensitive area information in the activation function layer,and further improve the ability of the network to extract features.?2?For the use of the softmax loss function for multi-class training,the extracted feature differentiation is not high,the model generalization performance is poor,and it is easy to misidentify.The additive angular margin loss function is introduced as the cost function of the model to expand the difference between the vein image classes of different fingers and reduce the difference between the vein image classes of the same finger,thereby improving the accuracy of finger vein recognition and enhancing the generalization performance of the model.At the same time,in the preprocessing stage of the finger vein image,the data augmentation method is used to further expand the finger vein image set and increase the training of finger vein image acquisition,thereby improving the robustness of finger vein recognition.?3?Aiming at the problems that the real-valued features extracted by the neural network occupy too much space and the retrieval speed is slow,the supervised discrete hash algorithm is used to discretize real-valued features,which greatly reduces the feature template.Using Hamming distance instead of Euclidean distance to measure similarity greatly improves the retrieval speed of finger veins.In order to verify the effectiveness of the algorithm in this paper,corresponding experiments are designed.The experiments will be performed on two public finger vein image datasets,MMCBNU₆000 and FV-USM.The algorithm has achieved good results in terms of finger vein recognition accuracy,retrieval speed,and template size,as demonstrated by experimental results.