Research On Dorsal Hand Vein Recognition Algorithm Based On Multiple Feature Points

How to safely and efficiently recognize an individual is an important problem that must be solved in the information age.The traditional recognition methods,such as ID cards and user names,are increaseingly difficult to meet the needs of society due to their own shortcomings,such as easy loss and leakage.In this context,biometric technology came into being.Dorsal hand vein recognition is an emerging biometric technology and it has many unique advantages,such as the internal body characteristics,living body,and non-contact.Because of this,dorsal hand vein recognition has increasingly become the focus of attention.Feature extraction is a key step in the dorsal hand vein recognition.In the images,the end points and the intersection points of the vein is often extracted as the feature information.Their extraction is relatively simple and can also reflect the structural information of the dorsal hand vein network.However,due to the limitations of image acquisition and other factors,the number of the end points and the intersection points that can be extracted is often too small,which makes the accuracy of the recognition algorithm difficult to satisfy.In view of above,combined with the images we collected,we have designed a set of feasible dorsal hand vein recognition algorithm and compared with algorithms that only use the end points and the intersection points,the accuracy of recognition was improved.The main work of this paper includes:(1)An image database of the dorsal hand vein was collected and built.Using the vein image acquisition device independently designed by us,380 images of the dorsal hand vein were collected,including 10 images of each dorsal hand of 19 individuals.The 10 images were divided into 5 images collected at two different times to increase the influence of the placement of the dorsal hand and ambient lighting on the algorithm.(2)Combined with the actual situation of our images,a full set of image pre-processing process was designed,and an ROI extraction algorithm based on the largest connected component of the image was proposed.Our preprocessing process mainly includes image type conversion,ROI extraction,image denoising,size normalization,grayscale normalization and image enhancement.During ROI extraction,several main-stream ROI extraction algorithms were analyzed,which were difficult to meet the actual requirements.Therefore,we proposed our ROI extraction algorithm,and which can effectively extract ROI.(3)A dorsal hand vein recognition algorithm based on multiple feature points was proposed.The maximum curvature algorithm was used to segment the dorsal hand vein images,and after that,vein repair,vein thining and deburring were performed.Due to the number of the end points and the intersection points was too small,we further extracted SURF feature points and used these three feature points as the feature information of our images.Then the extracted feature points are matched by Hausdorff distance and Euclidean distance,respectively.(4)Experiment and analysis.Integrated all the algorithms into the designed GUI.The EER of our algorithm based on Hausdorff distance is 1.58%,the single matching time is 0.27 seconds,the recognition rate is 96.32%and the single recognition time is 1.21 seconds.The EER of our algorithm based on Euclidean distance is 1.05%,the single matching time is 0.32 seconds,the recognition rate is 98.42%and the single recognition time is 1.81 seconds.Using our image database and experimental environ-ment,four comparative experiments were conducted.The results show that adding SURF feature points as feature information can effectively improve the performance of the algorithm,and compared with common recognition algorithms,our algorithms have better performance.