Research On Key Technologies Of Finger Vein Image Recognition

Finger vein recognition as a biometric recognition technology,its implementation principle is to obtain the vein image according to the infrared light of a specific wavelength,and extract the vein image features for individual identification and authentication.Because of its implementation principle,vein recognition technology is a non-contact biometric identification method.Facing the current COVID-19 environment,the authentication technology can effectively avoid the cross infection in the case of effective identification of individual recognition accuracy.At the same time,compared with other recognition technologies,finger vein recognition also has the advantages of living recognition,high accuracy and fast recognition speed.This paper deeply analyzes and explores the key technologies from the field of image preprocessing,image enhancement,feature extraction and recognition,so as to continuously improve the accuracy of the recognition algorithm.The main topics are as below:1.Firstly,the thesis made an in-depth study on the preprocessing part of finger vein.Based on the previous vein image noise removal and processing,an effective region of interest(ROI)extraction algorithm is proposed.Then the image size and gray level are unified by normalization algorithm.Finally,the vein image is segmented by Valley algorithm,Get the global vein features.2.From the perspective of improving the recognition rate of the algorithm,this paper studies the global feature extraction of images,and adopts machine learning algorithms to identify vein images in the image recognition stage.In the process of local feature extraction,a block LBP feature extraction algorithm based on weight allocation is proposed.And finally SVM was used to classify LBP features for training.Compared with other traditional global feature extraction algorithms such as HOG,SIFT and SURF,the experimental accuracy obtained on the relevant datasets is as high as 99.3%,and the accuracy rate is significantly improved.3.In order to further improve the recognition speed of the algorithm,this paper explores the current mainstream deep learning algorithm in the image field,first extracting the traditional LBP features of the vein image,then using the improved VGGNet network to train the LBP features,and finally using the Softmax layer to make multi-classification prediction of the output of the neural network.Experimental results show that compared with the traditional feature extraction algorithm,the deep learning algorithm has significantly improved the recognition speed,which fully verifies the effectiveness and speed of the convolutional neural network algorithm for finger vein feature extraction.4.Finally,aiming at the above algorithms,this paper develops the finger vein recognition system by using MATLAB tool software and MFC.Through the image user interface,the processing process and recognition process of the algorithm can be observed in real time,which provides a good implementation platform for further optimization and integration of vein recognition algorithm.