Research On Finger Vein Recognition Based On Shearlet

Because of the development of society and economy as well as daily updated high technology,human's living style has been greatly changing.Meanwhile,accurate,convenient and safe ID authentication technology is getting more and more attentions.Among various authentication technologies,biometrics identification is the most prominent one.These years finger vein has become more attractive due to some critical advantages,e.g.in-vivo recognition,high anti-counterfeiting,high acceptability,and high stability,etc.Finger vein recognition technology suggests a great prospect and huge potential.Currently,however,finger vein recognition has some problems as below:1,finger vein images are prone to influenced by the noise;2,the structure of some finger vein images is simple;3,finger vein images are prone to influenced by the unstable illumination;4,some information would lose due to irregular operations(cause great translation)when acquiring finger vein images.It is difficult for traditional methods to effectively handle these problems.As a novel multiscale geometric analyzing method,the Shearlet transformation is good at sparse representation and non-linear approximation expression,which enable Shearlet to extract structural information from poor-quality images.Therefore,on the foundation of the Shearlet transformation,this thesis conducts research in finger vein recognition from two angles:image denoising and feature extraction.The main work and innovation achievements of this thesis are as below:1.Based on zero mean SaS distribution,one Bayesian denoising algorithm in Shearlet domain is proposed to remove noises in finger vein images.Traditional denoising algorithms like spatial denoising and threshold-based denoising in transform domain denoising are prone to lose some information while removing noises,which usually leads to images degrade at different levels or incorporated some new artifact disturbances.The proposed denoising method in this paper is based on priori statistical models,which selects zero mean SaS distribution as the priori statistical model of one dimensional distribution of Shearlet sub-band coefficients of finger vein images,and then using Bayesian algorithm with MMAE to estimate those sub-band coefficients without noises,and finally reconstruct those estimated coefficients to get clean finger vein images.Experimental results show that:compared to traditional denoising algorithms,the proposed denoising algorithm in this paper can get better denoising results,because it can keep original structural information in images while denoising,which is beneficial to subsequent feature extraction.2.To resolve problems that finger vein has only simple structure and is prone to influenced by unstable illumination,one method of fused feature extraction based on discrete separable Shearlet(DSST)is proposed.Traditional feature extraction methods usually do not make full use of structural information of finger vein images and are prone to influenced by unstable illumination.Our feature extraction methods based on DSST has a great advantage that Shearlet can completely obtain singularity information of image curves.Feature extraction is conducted from three aspects:DSST coefficiency values,one dimensional distribution characters of DSST sub-band coefficients,and two dimensional structural characters of DSST sub-band coefficients.As coefficient characters in transform domain,DSST features are not only able to completely represent the structural information of finger vein images,but also not sensitive to pixel value fluctuation caused by unstable illumination,so DDST features are pretty robust.Additionally,our feature extraction methods do not include binary operations but decompose gray images and extract features directly,so they can avoid inverse influence of fake information caused by binary operations.A large amount of experimental results suggest that DDST fused features can extract finger vein features effectively especially for poor-quality finger vein images.3.In order to resolve the problem that some information would lose due to irregular operations(cause great translation)when acquiring finger vein images,NNST is proposed which invariant to translation.Traditional feature extraction methods usually do not make full use of structural information of finger vein images,thus could not get reasonable recognition results,especially when finger vein images translate too much and need to be cut a lot.However when applying Shearlet to extracting features of ROI,the image cutting problem(information lose)are converted to the ROI translation problem.What is more,NSST can completely get structural information of images and is resistant to unstable illumination and is invariant to translation,as a result,the coefficient features of NNST get more robust to overcome ROI translation.To improve the recognition rate of such a kind of finger vein images(some information would lose and/or image cutting caused by great translation when acquiring finger vein images),this paper extends the training sample set base on the improved ROI extraction method,and uses the improved robust regression for classification.Experimental results show that compared to traditional methods,our finger vein recognition method based on NSST do better in recognizing finger vein images which lack some information,so that they can guarantee effective recognizing in some specific situations.