Research On Dorsal Hand Vein Recognition Algorithm Under Partial Occlusion And Blur

With the development of artificial intelligence,biometric recognition technology based on machine vision has developed rapidly.Dorsal hand vein recognition has become an important branch of biometric recognition technology due to its uniqueness,stability,contactless and liveness detection.The existing hand vein recognition algorithms are all tested on clear vein images collected in ideal environment.However,in practical application,most of the real-time collected vein images have the problems,such as partial occlusion,blur and so on.The reasons are as follows:(1)the vein images are partially occluded due to pigmentation,scars,hairs or tattoos;(2)the relative motion between the palm and image sensor leads to the image blur.Feature discrimination ability of partial occluded and blur dorsal hand vein images is insufficient,which greatly affects system performance.This thesis takes the low-quality vein images as research object and focuses on the image acquisition device construction,region of interest(ROI)extraction,and robust vein feature extraction under partial occlusion and blur.The details are as follows:(1)This thesis has constructed vein image acquisition devices,and published two vein image databases.To improve stability of ROI,this thesis improves the circular ROI extrac-tion algorithm based on the Voronoi diagram.Delaunay triangulation is used to replace the conventional Voronoi diagram to locate inscribed circle,the largest inscribed circular ROI can adaptively adjust its radius as the palm contour changes.With the same feature,the recognition rate on our proposed ROI is 0.45%higher than the ROI extracted by Voronoi diagram.(2)Partial occlusion in vein images will lose some local vein information and affect system performance.This thesis proposes an improved biometric graph matching algorithm(IBGM)to extract features from partial occluded vein images.The algorithm adds the edge information to biometric graph registration and matching to improve discrimination ability of topology features.The accuracies on NHV(Normal Hand Vein)and THV(Tattooed Hand Vein)reached 98.03%and 97.14%,respectively.And the accuracy on artificial partial occlusion database reached 96.58%.Experimental results indicate that IBGM can effectively improve system performance against partial occlusion.(3)The textual elements degradation in blur image greatly affects system performance.This thesis proposes a multi-scale overlapped block local phase quantization algorithm.The algorithm divides ROI images into overlapped blocks,and uses short-time Fourier transform to extract multi-scale local phase information,which can increase feature representation ability and improve system performance.The accuracies on NHV and WHV(Whole Hand Vein)achieved 99.57%and 98.07%,respectively.And the accuracies on artificial blur databases reached over 99%(blur parameters<3),which is better than other algorithms,indicating that the proposed method can significantly improve the recognition performance against image blur.(4)To extract robust vein feature under partial occlusion and blur,this thesis proposed a RootSIFT key point extraction and matching algorithm based on image layers.This algorithm uses difference of Gaussian to extract occlusion and blur invariant key points.Firstly,the vein image is enhanced by the contrast limited adaptive histogram equalization algorithm with different sub-block,and the 10-layer enhanced image is constructed;then,the RootSIFT key points are extracted from each layer,and the intra-layer feature points are matched on the corresponding layers of enhanced template image and enhanced to-be-matched image;finally,all the matching point paires are gathered together,and the matching point pairs are filtered by topological relationship between feature points to improve the recognition performance against partially occluded and blur images.The accuracies on NHV and WHV reached 99.30%and 98.85%,respectively.On the artificial partial occlusion-blur database,the accuracies reached more than 95%.The results show that the algorithm can improve system performance against partial occlusion and blur.In summary,this thesis focuses on dorsal hand vein recognition algorithm under complex interference,and proposed ROI extraction and robust vein feature extraction algorithms un-der partial occlusion and blur,and these algorithms are applied to vein recognition system to improve its performance.