Research On Key Algorithms Of Finger Vein Identity Recognition

Finger vein image recognition now occupies an increasing proportion in the field of biometric recognition.Compared with fingerprint recognition and face recognition,finger vein image recognition has higher security due to its characteristics of vividness,high stability and non-contact.It has attracted extensive attention of researchers in various fields and has great research value.Aiming at the problems of low feature utilization,long time-consuming or low stability and the validity of classifier design in finger vein image recognition,three commonly used finger vein databases such as SDUMLA,polyU and MMCBNU,and self-built database are studied in depth.The main research contents and innovative achievements are as follows:(1)Aiming at the problem of image blurring and over-brightness caused by improper finger placement or improper pressing force in image acquisition process,a method of finger vein image evaluation based on multi-information fusion is proposed,taking self-built database as the research object.Planning image problem area,extracting brightness,information entropy and fuzzy variance features,and combining support vector machine(SVM)to get evaluation results.Experiments show that the classification accuracy of this method is higher than that of single feature or double information fusion feature,and the accuracy is 99.06%.(2)Aiming at the accuracy,success rate and efficiency of ROI extraction,a personalized ROI extraction method based on self-built database is proposed by combining Canny operator and quadratic segmentation strategy.First,combining the two-dimensional gradient Gaussian kernel and the idea of Canny algorithm to detect image edges.Then the edge binary image is projected horizontally and vertically,and the images are roughly segmented by finding the peak points of the planned region.Finally,the ROI after fine segmentation by removing the edge region.Experiments show that the success rate of ROI extraction from 144 images is 99.31% and only takes 30 ms,which is about 15% higher than that of gray histogram and region growing method,and the segmentation effect is more accurate.(3)Aiming at the validity and time of feature extraction and the “aliasing effect” of the histograms of orientation gradient(HOG)features,an improved HOG descriptor(ng_HOG)is proposed,which combines neighborhood gradient information.In the process of constructing the bottom histograms of orientation gradient,the neighborhood of the cell is planned,and the original HOG features are modified by the weights of distance and gradientamplitude.Experiments show that the recognition rate of this method is 98.23%,98.26%,94.18% and 97.62% on MCBNU,SDUMLA,polyU and self-built database respectively,and the running time of the algorithm is 224 ms,219 ms,117 ms and 57 ms respectively..(4)Aiming at the validity of threshold design of dichotomous classifier in illegal user recognition,illegal user recognition method of finger vein based on residual distribution is proposed by combining principal component analysis(PCA),sparse representation and Softmax function.According to the weight distribution of the reconstructed residuals after sparse representation and the probability distribution characteristics of image category attributes,the discriminant conditions of illegal user classifier are designed to identify the category attributes of legitimate users.Experiments show that in the experiment of illegal user identification on SDUMLA database,the correct rate of the method is 99.27%,the false recognition rate is 0.0002%,the rejection rate is 0.73%,and the average recognition time of a single image is less than 130 ms.