The Key Technology Reserch Based On Blurred Finger Vein Image

With the rapid development of information technology in recent years,the issue of personal information security has attracted more and more attention.How to correctly and quickly identify personal identity has become an urgent problem to be solved.Biometrics recognition technology is a fast identification method.However,traditional biometrics recognition technology is difficult to meet people’s high requirements for information security due to its damage easily and poor availability.Finger vein recognition technology,which uses the unique capillary lines inside different fingers as recognition features,has unique advantages such as in vivo collection,non-contact,high anti-counterfeiting and other unique advantages,which effectively solves the problems of low security faced by traditional biometric technology,and has become an indispensable way of identity authentication.However,due to the excessive pressure placed,equipment instability,such as excessive exposure or defocus of infrared camera,low temperature and other factors in the process of finger vein image acquisition,which will cause poor contrast and blurred vein texture in the collected finger vein image,lead to poor recognition performance.Therefore,this paper mainly explores the deblurring and recognition algorithms for blurred finger vein images.The following is the main research content of this paper:(1)The paper analyzes the blurred finger vein images acquired by different placement pressures,different external temperatures,different camera exposure parameters and other actual scenes.According to the blur characteristics,they are divided into two types: local blurred finger vein images and global blurred finger vein images.The vein texture in the central area of the local blurred image is more blurred than the peripheral area,while the overall vein texture of the global blurred image is blurred.It is pointed out that the blur degree of the finger vein image caused by different degrees of blurring conditions is different,even if the different blurred finger vein images acquired by the same finger.(2)A finger vein image deblurring algorithm based on Neighbors-based Binary Generative Ad-versarial Network(NB-GAN)is proposed,and a 26-layer deep network generator constrained by NBP texture loss is used to restore clear images.Firstly,according to the characteristics of the blur types and the various blur degrees of the finger vein blur images in the actual application scene,a method of alternately iterating convolution defocus and mean blur kernels in a multi-scale window is proposed to produce a polymorphic blur training set.Then it is proposed to use Neighbors-based Binary Patter(NBP)texture loss to constrain the generator to generate high-fidelity finger vein images.Finally,a convolution mode with all convolution steps of 1 is proposed to retain more vein texture feature information,and residual jump connections are added on both sides of the residual module of the 26-layer deep network generator constrained by NBP texture loss.To prevent degradation and overfitting of the training model.Theoretical analysis and simulation results show that the proposed deblurring algorithm based on NB-GAN achieves better deblurring performance than the widely used Deblurring Generative Adversarial Network(Deblur GAN)algorithm.(3)A blurred finger vein image recognition algorithm based on multi-scale local feature fusion is proposed.This algorithm solves the problem of reduced feature accuracy caused by image blur by expanding the calculation scale and increasing the feature dimension.Firstly,according to the characteristics of finger veins,a multi-scale direction template is proposed to calculate the 8direction response values of the image.Secondly,in order to avoid the error caused by image blur,the neighborhood where the 8 direction response values are located is averaged and then compared,where the direction corresponding to the maximum value is the local direction feature.Then,the extracted 8 directional response values are calculated by the Multi-scale local binary pattern(MLBP)operator proposed in this paper to obtain high-dimensional local structural features to reduce the sensitivity of the features to pixels.Enhance the accuracy of features.Finally,the local structural features and local directional features are combined to match by the optimal weight method,and the direction and structural characteristics of the finger veins are fully utilized to further enhance the accuracy of the extracted local features.After sufficient experimentation,the algorithm proposed in this paper has better recognition performance on blurred finger vein images than NBP,Neighborhood Matching Radon Transform(NMRT)and Multi-block Mean Neighbors-based Binary Pattern(MMNBP).