Research Of Finger-vein Recognition Based On PDEs

With economic development and social progress,the traditional authenticationmethods have been unable to meet the demand for high security. Under suchcircumstances,biometrics emerged accordingly.Biometric identification technology isthe technology to implement personal identification by using the inherentphysiological characteristics of human body. The finger vein recognition technologyis a new contactless biometric technology, due to its high accuracy, stability,non-aggressive and unigueness,it gradually becomes the current hotspot.However,this technology started relatively late in the mainland,there is nomature products on the market currently.So the study on the finger vein recognitiontechnology is not only conduce to technological innovation,but also can speed up thepace of getting rid of rely on foreign technology companies,it has great significance.This thesis is supported by the Science Technology Development Project of JilinProvince.Based on the latest extensive discourses and technology journals in thisfield,the thesis is trying to conduct in-depth research and study of the entiresystem.We apply the theory of partial differential equations to the field of finger veinrecognition and achieve some results.In this thesis,mainly from the following aspects:First of all,the background and the meaning of this research are given.Desicribethe relevant theory of the finger vein recognition technology and review developmentsituation domestically and abroad.Summarize the application and development ofpartial differential equations in the field of image denoising and segmentation.Theevaluation criteria and theoretical basis of the partial differential equations involved inthis thesis are described in detail.Secondly,the characteristic of infrared finger vein image is studied,the fingervein images we obtained usually contain severe noise, shadow,polarization and theproblem of low contrast. Implement the image background removal andnormalization,and complete the pretreatment in this work.Introduced two newdenoising modals based on partial differential equations,they can maintain the edge ofimage while denoising. One modal utilizes a fourth-order PDE model to improve thesecond-order PDE, and can increase the peak value Signal-to-Noise Ratio (PSNR) up2dB. The other one uses the new diffusion function based on the traditional P-M model and in theSignal-to-Noise Ratio is improved.Then,this thesis discusses a variety of segmentation algorithms and the thinning method is given in the stage of the feature extraction.A midpoint thresholdsegmentation method is proposed in this work,and it is an improved multi-thresholdalgorithm can split the finger vein image quickly and clearly.Based on the activecontour model and its relevant theory,in this thesis we present two segmentationmodels based on partial differential equations.One uses the local binary fitting item toobtain optimal evolution in the global scope, and uses the additional gradient item andthe pullback force item to let the evolution of the curves which in the smooth regionsor deeply concave regions, stop in the target boundaries more accurately andrapidly.The other one is an active contour model with mixed information based ondouble-well potential initialization,which introduces the double-well potentialinitilization based on mixed information model to ensure the stability and rapidity ofthe curve evolution.A large number of experiments proved the effectiveness of thetwo new models.Finally,in the stage of matching and identification,this article implements arecognition algorithm based on finger vein structure,and the matching curve and ROCcurve are given. Experiment results prove that this algorithm can overcome the issue ofsmall deformation and rotation of the template.