Finger Vein Recognition Algorithm With Anti-Rotation Property

With the increasing awareness of information security,more and more attention has been paid to biometrics.Among them,finger vein recognition that takes full advantage of its liveness detection property,convenient operation,low cost,has been highly approved by the market.However,because of the non-fixed position during the image capture,the impact of rotation around the finger is always a serious problem which may lead to deformation and translation.Aiming at this rotation problem,this paper has explored the solution from multiple perspectives.Compared with the existing researches,the key contributions of this work can be summarized as follows:First,a new rotation correcting method is proposed.Based on the analysis of the rotation problem,we found it will lead to the image transformations,such as the finger vein deformation and translation,the change of background brightness distribution,as well as the unexpected disappearance and appearance of some finger veins near the edge.Among them,we compared the impact of deformation and translation,and discovered that translation is the main issue resulted in performance degradation.Therefore,a new rotation correcting method based on finger vein gradient is proposed.Experimental results showed that this method is effective at reducing the impact of small-angle rotation.Second,a soft biometric with anti-rotation property is given.Based on our analysis,it can be found that the finger vein image consists of not only finger vein network but also the features of background brightness distribution.Compared with finger vein network,background brightness distribution is more robust against rotation around the finger.Therefore,we proposed a soft biometric based on background brightness distribution.This method extracts the finger vein background layer,in which the statistics information is used for describing the brightness distribution.The fusion of this soft biometric and main biometric can reduce the impact of rotation.Third,a mixed match method for main biometric and soft biometric is presented.The main biometric and soft biometric must be matched with different methods respectively because of their difference in dimension and magnitude.To address this problem,we proposed a mixed match method,in which support vector machine(SVM)is applied to match the high-dimensional main biometric and Manhattan Distance is used for matching the low-dimensional soft biometric.This match strategy includes two benefits: better performance of classification for main biometric with the optimal hyperplane trained by SVM,and effective match for soft biometric with the low computational cost of Manhattan Distance.Finally,the equal error rate of our methods in three open access databases(FV-USM,MMCBNU and SDUMLA)are 0.216%,0.827% and 0.712% respectively,which demonstrated that our methods can not only improve the system recognition performance but also increase the capacity of anti-rotation.