Research On Feature And Classifier Of A Large Sample Of Dorsal Hand Vein

Dorsal hand vein recognition is a newly developed biometric method in recent years, and also become one of the focuses in the field of biometrics authentication. The hand vein features uniqueness and universality. The venous vascular network is not the same between everyone, and even the between the hands from the same person, but the vascular structure of the individual does not change with time. Because of hand vein is located in the body, it is not easy to be affected by pollution, scratches and other external factors. Moreover, it is not easy to be imitated. The vein image acquisition process in hands with respect to other parts of the body is much more acceptable. All these properties make the hand vein a qualified and stable biometric characteristics.Facing the concrete situation of China’s large population and the frequent movement of personnel, it is necessary to study the key problems of dorsal vein identification on large sample condition. Compared with the success of fingerprint, face and iris recognition technology, how to achieve robust identification in the condition of large samples of hand vein, is of great significance to carry out theoretical research and applied research related scientific research work to promote the hand vein identification technology for further.This paper mainly focuses on the problem of the hand vein identification technology on large sample size, especially the study of designing dynamic update classifier, and the optimal local descriptor of near-infrared vein structure. The innovative works of this paper are show below.(1) Proposes a novel segmentation method of hand vein image. The adaptive gradient separation algorithm based on self adaptation to less or over segmentation compensation, has better performance compared with other classical algorithms.(2) Proposed a SIFT feature matching method for eliminating the error. With the help of many-to-one matching method, horizontal slope matchline detection and vertical slope detection, two feature points, only with a similar description vector and a similar coordinate position, will considered correctly matched. This ensures the correct match rate of SIFT feature points.(3) Proposed a SIFT feature based template recognition algorithm. SIFT algorithm for the feature extraction of redundant information contained a lot of problems. Start from the intra class and inter class relationship, the SIFT feature points are reorganized as union, intersection and exclusion set class. Because of the different expression of different point combinations, they have complementary information. Experiments show that the fused classifier template can achieve nearly100%recognition rate. This ensures the recognition accuracy for a large sample, and when adding new samples, the classifier will be updated by adding the new category to the classifier.(4) Proposed the MB-CSLBP local descriptor. MB-CSLBP description operator combines MB-LBP and CS-LBP to their respective advantages, averaging algorithm with large area, which makes the image more insensitive to noise. This operator respectively from the macro and micro perspective of the pixels are described, so that the two kinds of feature information is complementary, the overall characteristics of better highlight image.