Research On Robust Feature Extraction Of Dorsal-Hand Vein For Weak Constraints

Nowadays,most researches of dorsal-hand vein recognition are under conditions with high cooperation of users and single acquisition equipment.But,with the fast development of Internet,there exist many weak constraints,such as different acquisition equipments,different environments,different postures of hand and low cooperation of users.These weak constraints can change the quality of dorsal-hand vein images,such as contrast,rotation,affine deformation and so on.Meanwhile,they can reduce the recognition rate of dorsal-hand vein.So,how to identify the dorsal-hand vein accurately and effectively under weak constraints still needs further research.Based on the dorsal-hand vein database under weak constraints,the main innovations in this paper are summarized as follows:(1)Researched different types of dorsal-hand vein images(gray,binary and gradient);Analyzed different methods of key points extraction(Gaussian random,concentric circle and Difference of Gaussian).These different image types and different methods of key points extraction can form nine compound modes.In the case of SIFT(Scale-invariant feature transform)descriptor which has 128 dimensions feature descriptor,experiments showed that binary images and different of Gaussian can make the recognition rate of dorsal-hand vein under weak conditions up to 82.6%.(2)Optimized the SIFT of feature extraction and matching aimed at the characteristic of dorsal-hand vein.Optimizations included four aspects:Chose the best scale vector ? 1.0;Chose the new extreme points search template with the size of 5×1×3;Changed the all extreme points to the minimum points;Chose the best matching threshold R 0.88.After the optimization of SIFT,with the binary images and different of Gaussian,the results of experiment showed that the recognition rate of dorsal-hand vein was up to 89.43%.(3)Designed a Distinctive Efficient Robust Feature from the biological modeling of P Ganglion cells.This descriptor regarded the Gaussian Function as the convolution kerned and had exponential scale and sampling grid structure.Compared with the SIFT descriptor which has 128 dimensions feature vector,the new descriptor which has 328 dimensions feature vector could describe the feature more accurately.Combined with the optimized feature extraction and matching and the binary images,the experiments showed that the recognition rate of dorsal-hand vein was up to 90.47%.In a few words,the innovations made the recognition rate of dorsal-hand vein from 82.6%up to 90.47%.It proves the effectiveness of our work.