Research Of Finger-vein Feature Extraction And Anti-spoofing Detection Based On Deep Learning

Along with the coming of the information age,the demands for security are becoming more and more important,and people pay a high attention to the security of personal products,personal finance,and e-commerce.However,identity authentication is the premise of ensuring information security.Biometric identification,as an identity authentication technology,has attracted more and more attention because of its high security and convenience.With the development of society and progress of technology,its application has become increasingly extensive,such as access control systems,ATM systems,security monitoring systems,health care and information security.Among the many biometric identification technologies,finger vein recognition technology is becoming one of the most promising technology in recent years because finger vein conceals body interior,and is not easily copied and forged.In recent years,deep learning technology has been successfully applied in computer vision,natural language processing and so on,and achieved a series of achievements.Based on this,this paper combines the deep learning theory to carry out a series of research on feature extraction and anti-spoofing detection of finger vein images.The main research contents can be concluded as follows.(1)We proposed a finger-vein image segmentation approach based on sparse selfencoder.First of all,this paper proposes an automatic labeling method to label veins and the background pixel.Then,training sets is created which is for training.Finally,the trained model are used to segment the test images.This paper carries out experiments on public data.The experimental results show that the proposed algorithm is superior to the ordinary segmentation algorithm based on manual feature,which effectively reduces the error rate of the finger-vein recognition system.(2)We proposed a finger-vein feature restoration approach based on full convolutional neural network.First,the original image is segmented by sparse self-encoder to obtain a binary image,and the binary image is refined to extract thinned image.Then creating training set to train the full convolutional neural network model.The trained model is used to restore the vein patterns to obtain a more complete finger vein feature.Finally,the vein minutiae points are extracted to achieve the authentication of the individual identity.The experimental results on the public data show that the proposed algorithm not only can restore the vein patterns,but also improve the recognition performance.(3)We proposed a finger-vein image anti-spoofing detection approach based on deep belief network.First,a finger vein image is divided into different small block to establish training set.Then,training the model and the trained model is used in the identification of spoof finger vein images.The experimental results on the public data show that the proposed algorithm can detect the spoof finger-vein image,and effectively improve the recognition accuracy of the finger-vein recognition system.