Font Size: a A A

Research On Finger Vein Recognition Algorithm

Posted on:2020-07-12Degree:MasterType:Thesis
Country:ChinaCandidate:Y Y PanFull Text:PDF
GTID:2428330626453419Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
Finger vein recognition technology has the characteristics of uniqueness,stability,safety,simplicity and ease of use,which makes it stand out in other biometric technologies.However,the recognition rate of finger vein recognition system is vulnerable to translation,rotation,ambiguity and low contrast of vein images.In order to solve the problems,this thesis studies the finger vein recognition system from three aspects: finger vein image enhancement,finger vein image registration and finger vein recognition.The main contributions of this thesis are introduced as follows:(1)An enhancement algorithm for the finger vein image is proposed.First,all possible vein regions are extracted by the local features of the image,and then the filtering operation is performed on the image by using the pilot filtering.Experiments show that the enhancement method of this paper can effectively improve the correct rate of finger vein recognition.The recognition rate of SDUMLA-FV in Shandong University is 93.2%,and the recognition rate on the integrated data set is 83.6%.(2)The existing single slider registration method is improved by using the convex hull of image vertical projection histogram to locate finger joints.The time complexity of the improved method is lower than that of the original method,and the accuracy of the improved method is also improved.(3)The fine-tuned AlexNet is used for finger vein image registration.Using the open source tool LabelMe to manually mark the joint point coordinates,and then put the marked data set into the fine-tuned AlexNet network for training.Experiments show that the deep learning method has higher precision than the traditional method.The recognition rate of the image after registration is 13.2% higher than that of the unregistered image.(4)A finger vein recognition method based on Hessian-Affine operator is proposed for small sample method.The Hessian-Affine feature has high robustness to illumination,translation,rotation,scale and other transformations of vein image.In the 1:N recognition experiment,this method has an accuracy of 4.71% higher than the traditional SIFT recognition.(5)A finger vein recognition method based on Hessian-Affine feature operator is proposed for large sample problem.In order to improve the robustness of the model against illumination,translation and rotation,The Center Loss assisted Softmax cross entropy is used as the loss function to increase the distinction between feature vectors,and add regular termsto reduce the over-fitting of the model.The recognition rate reaches 99.5% on the comprehensive data set composed of four open data sets.
Keywords/Search Tags:Finger vein image enhancement, finger vein image registration, Hessian-Affine, ResNet
PDF Full Text Request
Related items