Research On Sample Generation And Recognition And Anti-spoofing Integration For Finger Vein Authentication Systems

With the improvement of social economic information and the enhancement of security awareness,biometric based identity recognition has attracted more and more attention.With the development of deep learning,the emerging finger vein recognition,with its high security,high affinity and unique live detection features,has been applied in high security needs,including entry-exit management,financial services,e-commerce,information security and other occasions.However,there are two problems existing in the development and application of finger vein: the lack of data samples and the axial rotation of fingers,which seriously limit the accuracy and robustness of deep learning based algorithm.In addition,the existing system uses independent algorithm model to process the finger vein recognition and anti-spoofing tasks,which reduces the efficiency and real-time performance of the system.In view of the above problems,this paper first uses the adversary generation network to augment the data,and uses the generated samples to improve the finger vein recognition performance.On this basis,it further studies the integration of finger vein recognition and anti-spoofing task.The main work of this paper is as follows:First of all,the current deep neural network model in finger vein recognition is difficult to further improve its recognition accuracy and generalization performance due to the small data size and the lack of finger axial rotation samples.Therefore,this paper proposes a data generation method based on the adversary generation network,and designs an efficient adversary model and training framework for the existing finger vein image from the perspective of inter class sample and intra class rotation sample generation.The rotation sample generation model proposed in this paper can effectively learn the rule of the influence of finger axial rotation on the vein image,and generate high-quality samples according to the vein image in normal posture to augment the data,so as to further use in the existing deep model training and improve the robustness of the algorithm.Then,in the finger vein recognition system,recognition and counterfeiting detection are usually separated as two subtasks.In order to further improve the efficiency of finger vein system,this paper constructs an integrated recognition and anti-spoofing model FVRAS-Net based on the deep convolution network to handle two tasks at the same time.The trainingframework is designed by the way of multitask learning to effectively extract the features of with both stage unified.The model designed in this paper can extract the features of recognition samples and anti-spoofing samples at the same time,which can effectively ensure the accuracy of recognition and anti-spoofing,and improve the efficiency and real-time of finger vein system.Finally,in order to effectively evaluate the performance of the proposed algorithms,experiments are carried out on multiple open datasets and self built datasets.The final results show that the accuracy of the recognition model and the robustness to the rotation samples can be effectively improved by the inter class and intra class rotation samples generated by the proposed model.At the same time,the anti-spoofing and identification integration model proposed in this paper achieves 100% anti-spoofing rate and 2.1% equal error rate on SCUT self built database.In addition,the anti-spoofing and identification integration model proposed in this paper has been deployed on the embedded system and can run stably offline,which proves the feasibility of the algorithm in this paper.