Research On Finger Vein Recognition Based On Rotation Correction

In recent years,due to its security,convenience,uniqueness and stability,biometric authentication technology has gradually replaced the traditional authentication methods.In particular,finger vein recognition has attracted widespread attention with its unique liveness capability,and been selected by many companies as commercial product development.However,in practical applications,the method of reducing finger restraint,used to improve the affinity of the system,lead to the high degree of freedom of the finger during finger vein acquisition.Especially,the axial rotation of the finger in image acquisition could cause the inconsistencies in the imaging area and pattern deformation,which would make great impact on the recognition performance of the existing algorithms.Considering the existing issues,the paper makes a more in-depth analysis and research about finger vein on the image preprocessing,feature extraction and coding,as well as feature matching methods.Compared with the existing researches,the contributions of this work can be summarized as follows:First,a novel feature,called double orientation coding histogram,is proposed.With the problems of the inconsistencies in the imaging area and pattern deformation caused by the axial rotation of the finger vein,this paper uses the vertical texture method to extract mutual texture area.However,the pattern deformation problem still exists caused by the axial rotation of the finger between the obtained mutual areas.To this end,this paper proposes a double orientation coding histogram feature to deal with the deviation and deformation of the veins in the mutual region,which can effectively reduce the impact on the axial rotation of the finger.Second,a feature extraction method based on the minimum convolution point and its corresponding deformable matching method are proposed.As mentioned above,after obtaining the mutual area,the inconsistent vein image and pattern deformation can affect the performance of the finger vein recognition.Therefore,this paper utilizes multiple convolution filters to enhance the original image and extract the feature points,and combines the new deformable matching method to deal with the above-mentioned pattern deformation problem.By comparing with the improved SIFT feature method,the validity of minimum convolution point feature is verified.On the basis of this,this paper further improves the performance of finger vein authentication through the fusion of texture feature extraction and achieve the best effect of current finger vein authentication.Third,a paired SVM authentication method is proposed.At present,the distance-based finger vein authentication cannot adaptively determine the authentication threshold,and manual selection of the threshold may affect the authentication performance.Therefore,by using the registration information of image layer and the difference information of feature level,we combine the mutual area textural information and SVM,and proposes a paired SVM authentication method to improve the performance of finger vein authentication.Finally,we conducted an algorithm evaluation of finger vein authentication and finger vein recognition respectively on the public dataset and the self-built dataset.The result of finger vein authentication experiment shows that the proposed fusion authentication method,based on the minimum convolution point and LBP,achieve the best performance in multiple database.In addition,the paired SVM authentication method also results in competitive performance on multiple database.The experiment result of finger vein recognition shows that the proposed method based on the minimum convolution point and its corresponding deformable matching method can also achieve the highest recognition rate in multiple database,which proves the effectiveness of the proposed method.