Research On Dorsal Hand Vein Recognition Based On Graph Matching

In the Internet era when identity recognition is getting more and more attention,identity recognition based on dorsal hand vein has become a research focus.However,in face of the diversity of collection device,identification based on dorsal hand vein under cross-device condition has become an inevitable trend.There are differences in brightness,rotation angle,and scale of dorsal hand vein images collected under cross-device and low user cooperation conditions,so recognition rate is low.How to carry out high-precision identification based on dorsal hand vein under cross-device condition has important research significance.The main work and innovation of this paper are as follows:(1)Propose two schemes to describe the shape structure of veins with skeleton image: Scheme One is to get skeleton image by thinning binary image,and Scheme Two is to get skeleton image by thinning distance image.On the basis of these schemes,select bifurcation points and ending points on vein skeleton as the key points,then intercept binary image patch(default is 48?48)with key points as the center,and finally use the DERF descriptor to generate feature vectors for dorsal hand vein recognition.The single-device recognition rate of library 1 and library 2 reached99.18% and 97.35% respectively with Scheme One,and 98.98% and 97.35%respectively with Scheme Two.The recognition rate under cross-device condition is71.43% with Scheme One and 70.41% with Scheme Two.Therefore,the recognition rate with Scheme One is relatively higher,and the follow-up work is carried out on the basis of Scheme One.(2)In order to improve the recognition rate under cross-device condition,propose an algorithm that takes bifurcation points,ending points,and points sampled at equal intervals from vein edges as the key points.Inspired by graphics,describe the vein skeleton by points and edges,so as to describe the shape structure of vein more completely.Optimize the sampling interval and the size of binary image patch centered on key points,and the optimal values are 7,60?60 respectively.Use the DERF descriptor to generate feature vectors.The recognition rate under cross-device condition has increased from 71.43% to 81.84%.(3)In order to further improve the recognition rate under cross-device condition,propose an algorithm for dorsal hand vein identification based on multi-image fusion and improved Xception network.At the network input,based on the idea of combining texture features and shape features,fuse binary image,distance image,and skeleton image into a three-channel fusion image.Inside the network,improve the Xception network,forcing the network to learn more sparse features,and the improved recognition rate reaches 93.54%.The efficiency of the above methods is sufficiently proved on the cross-device dorsal hand vein image library 1 and 2 built by the laboratory in this paper.