The Fusion Of Finger Vein And Finger Shape Recognition Algorithm With Anti-Rotation Ability

With the development of science and technology,people pay more and more attention to the security of their own identity information,and biometric recognition has also received more and more attention.Among them,finger vein recognition has attracted special attention of researchers because of its unique ability of living-body detection.However,in the process of finger vein image acquisition,the user's finger is in an unconstrained state with high degree of freedom,which is easy to cause vein image deformation caused by finger axial rotation and reduce the recognition accuracy of the system.The problem of finger axial rotation has always been a challenging problem.Therefore,this paper makes a deep research on this problem and explores the method to solve the problem of finger axial rotation from multiple perspectives.Compared with the existing research,the contribution of this paper is as follows:First,a new algorithm of finger vein feature extraction is proposed to resist finger axial rotation in the common region.Aiming at the problem that the axial rotation of fingers will lead to some inconsistencies in the imaging region,this paper first uses the common region extraction method based on Pearson similarity to obtain the most similar region of the image to be matched.However,there are still texture deformation between the most similar areas,which are caused by the axial rotation of fingers.Therefore,we propose a double orientation coding histogram feature to solve the common problems of vein texture deformation,so as to effectively reduce the impact of finger axial rotation.In addition,by analyzing the feature information contained in finger vein texture,we can find that finger vein texture contains not only orientation information,but also structure information.Therefore,the fusion of double orientation feature and structure feature is proposed in this paper.Experiments show that the fusion of the two features can effectively improve the recognition performance.Secondly,a new finger shape feature extraction algorithm is proposed.For the original vein image,we can see that the finger area in the image contains not only vein texture information,but also finger contour information.In this paper,finger shape features are used as auxiliary features,which are fused with vein features to improve identity information and recognition performance.In order to reduce the redundancy between finger shape information,this paper extracts the representative width vector and centroid contour distance vector as bas-ic finger shape feature,and then generates the array of mean and variance by block processing.Experiments show that the block processing strategy can not only greatly reduce the redundancy of information,but also alleviate the impact of finger axial rotation.Thirdly,a hierarchical hybrid matching strategy is proposed to match the finger vein and finger shape features.The amount of information contained in finger vein and finger shape features is different,in order to get better matching results,finger vein feature and finger shape feature should use different matching methods in the matching stage.For vein structure feature and vein orientation feature with more information,the complex SVM classification and chi-square distance are used to obtain their matching scores.Then the two vein feature scores are weighted and fused to get the final vein matching scores.For finger shape features with less identity information,Manhattan distance is used to get the final finger shape score.Finally,the final matching score was obtained by weighted fusion of vein score and finger shape score.The hierarchical hybrid matching strategy fuses the finger vein information and finger shape information,so that the final matching score has more identity information and improves the accuracy of identity recognition.Finally,this paper not only carries out experiments on the open transmission database FV-USM,MMCBNU and SDUMLA,and achieves the equal error rates of 0.11%,0.37% and0.46%.Moreover,experiments are also carried out on the self-built reflective database SCUT LFMB--D:RIFV,which achieves an equal error rate of 2.26%.According to the experimental results,the proposed method can not only improve the recognition performance of the system,but also effectively improve the anti-rotation ability of the system.To a large extent,it solves the problem of the decline of recognition performance caused by the axial rotation of fingers.