Research And Implementation Of Multi Finger Vein Recognition System

With the rapid development of mobile Internet,more and more account theft and property loss cases occur frequently.How to effectively and conveniently protect users’ information security and privacy has become the focus of large enterprises and research institutions.In order to ensure user information security and more convenient user identity authentication,finger vein recognition technology came into being.Finger vein recognition has the excellent characteristics of in vivo detection,high accuracy and not easy to lose.In the past,the research of finger vein recognition mainly focused on the field of single finger vein recognition.However,due to only collecting a single finger for finger vein recognition,there are some limitations in safety and success rate of single verification.As a kind of finger vein recognition,multi finger vein recognition has high research and application value by virtue of the vein area of multiple fingers.In view of the current research situation of multi finger vein recognition,this paper explores the combination of deep learning and multi finger vein recognition,and proposes a set of solutions for multi finger vein recognition.The specific research contents of this paper are as follows:(1)In view of the shortage of multi finger vein data sets,this paper uses multi finger vein collection equipment to collect multiple finger veins of the same person sample.A total of 19 character samples were randomly selected in this collection.Each character sample collected 10 finger vein images of the left and right hands at the same time.In order to reduce the accidental error and increase the richness of the sample set,each person sample repeatedly collects the multi finger vein image 50 times.Through post clipping and production,a total of 1900 images of multi fingered vein data sets were obtained in this acquisition.In addition,in order to improve the image quality of multi finger vein data set,this paper also carries out preprocessing and data expansion on the data set,which further improves the indicators of subsequent algorithms.(2)Aiming at the problem that the multi finger vein image contains multiple regions of interest and different directions of regions of interest,this paper proposes a multi finger vein region of interest detection algorithm mfvn based on rotating target detection.The algorithm takes centernet as the baseline and can accurately detect the region of interest in the multi finger vein image.In order to more accurately detect the orientation of the region of interest of multi finger vein,this paper uses the tan value of predicting the angle of the region of interest to improve the accuracy of angle information prediction.Finally,the mfvn network proposed in this paper has achieved good results in the region of interest extraction task of multi finger vein image.(3)Aiming at the limitations of single finger vein in safety and detection success rate,this paper proposes two modes of multi finger vein recognition:safety mode and high pass rate mode.Cosface loss is used as the loss function to train the corresponding models for the two modes respectively.Finally,experiments show that the accuracy of multi finger vein recognition is higher than that of single finger vein recognition.(4)Finally,based on the development architecture of front-end and backend separation,this paper develops a display system of multi finger vein recognition,uses the multi finger vein region of interest extraction algorithm and multi finger vein recognition algorithm proposed in this paper as the theoretical support of the system,and implements the two multi finger vein recognition modes proposed in this paper in the system.