Research On Key Technologies Of Finger Vein Identification

Nowadays,with the advancement of technology and the approaching of a cashless society,personal identity authentication and privacy security issues have become more and more important.More and more biometric-based identification technologies are beginning to enter our lives.In recent years,academia has proposed identification based on in vivo biological characteristics,and with the maturation of infrared imaging technology,pattern recognition,and artificial intelligence and other related technologies,it has become possible to make use of the human body's biometric finger veins as authentication features.At present,the conventional method of finger vein recognition technology is based on the image feature method,and its main idea is to extract features of the overall image or features of the vein pattern.Because there are a large amount of redundant data based on features acquired from the whole finger vein image,the time complexity is high,and the features extracted from the vein pattern are greatly affected by the image segmentation algorithm.In order to improve the accuracy of the finger vein recognition algorithm under small samples,A finger vein recognition algorithm based on deep belief network and convolutional neural network is proposed in this paper.First of all,this paper summarizes some of the machine learning theories involved in the paper,including deep belief network and convolution neural network.Secondly,from the research motivation,the experimental steps and the results analysis,the two new algorithms of finger vein recognition are described in depth:Finger vein recognition method based on deep learning and improved Gabor feature fusion and finger vein recognition method based on improved convolutional neural network.Finally,we summarize the work arrangements of this paper,and elaborate the future research priorities.Finger vein recognition algorithm based on deep learning and improved Gabor feature fusion: First,the traditional Gabor filter is modified to have a curvature response capability,and then a uniform LBP operator is used to further process the fusion feature,and finally the fusion feature information is input to the deep belief network performs vein image recognition.Finger vein recognition algorithm based on improved convolutional neural network: Firstly,this method increases the ability of the network to extract features by increasing the number of convolutional layers,and uses the improved activation function to improve the nonlinear capability of the network;secondly,an improved pooling model is used.When the effective features are retained,the network feature dimension is reduced,and Fisher discriminant information is introduced as a constraint condition when the adjustment weights are back-propagated.Experimental results from the finger vein library in Tianjin and the finger vein library(FV-USM)from the University of Science and Technology of Malaysia indicate that the accuracy of the finger vein recognition method proposed in this paper is better than traditional classical algorithms,and has good recognition performance and application value.