Adaptive Palm Vein Recognition Method

In recent years,palm vein recognition technology has been widely applied and its market share has continued to increase.However,most palm vein recognition methods still face issues such as low image utilization rate,image distortion caused by image scaling,and potential leakage of user information.Currently,most methods for extracting the region of interest in palm vein images are based on locating the valley points of the fingers,which is prone to failure when processing non-contact datasets,leading to low image utilization rate.Furthermore,existing deep learning models for palm vein feature extraction require input images to be scaled to a consistent size,which can lead to image distortion due to the area of interest in different palm vein images has different sizes.Additionally,current palm vein recognition devices typically store both user identity information and biological feature information locally,increasing the risk of user privacy information leakage.To address the aforementioned issues,this paper proposes an adaptive palm vein recognition technology and an identity authentication system based on the fusion of palm print and palm vein features.The innovation of this paper mainly lies in:Firstly,a palm vein image region of interest extraction method based on maximizing the inner tangent circle of the palm is proposed in this paper.By locating the centroid,the method can effectively extract the region of interest in non-contact palm vein images.Using image erosion and rotation correction assistance,the region of interest can be maximized and has rotation robustness.In a dataset with an image resolution of 800*600,an region of interest of average size 352*352 can be extracted.Secondly,an adaptive palm vein feature extraction model is proposed,which expands the training and testing data through dataset fusion and augmentation.The structure of the network model is improved to accept inputs of any size,and the loss function and normalization strategy are improved to enhance model performance.Contrastive experiments using different data augmentation parameters,network structure parameters,and loss function weights show that the best equal error rate is 0.29%,proving the effectiveness of the model.Thirdly,a palm print and palm vein feature fusion algorithm based on eigenvalues is proposed,which uses different parameters to combine and contrast the eigenvalues.Compared with the original single-modal model,the recognition rate is improved by 17%.Fourthly,an identity authentication system based on palm print and palm vein feature fusion is designed and developed.Considering user usability and information security,palm print and palm vein images are collected in the same period,and the user’s identity information and feature information are unbound through cloud servers and local terminals.Equipped with the palm print and palm vein feature fusion algorithm,the system can effectively identify and store user identities.