Research Of Finger-vein Recognition Algorithm

With economic development and social progress, the security authentication products based on of biometrics technology become to serve the public. At present the entire biometric market are rapid developed and have a vast reservoir of potential. The adapting to market demand biometric identification technology and equipment of independent R & D, not only can promote the development of related domestic industries, but also conduce to technological progress and innovation, master core technologies: from the image acquisition equipment to image analysis, to biometric recognition algorithm. By this we can get rid of rely on foreign technology companies. This will ultimately benefit the prosperity of bio-certified industry and promote the international competitiveness. Fingerprints as biometric characteristics of certified products are firstly widely used, but there are also drawbacks, for instance, the security is not high, fingerprint characteristics are instability, the technology is not universal. As compensation, vein distribute in the body, the external environment will not be affected, there is no possibility of being imitated, precisely the attributes of human vein can overcome the shortcomings of the fingerprint. Among veins, the characteristics of a finger vein have a greater advantage: finger vein's collection equipment is small, the credibility of certification is high, everyone has ten fingers, and the sources of information are adequate. Finger vein recognition has gradually become a new field of biometric technology research.This paper studies the finger vein segmentation, feature extraction and matching algorithm and the algorithm to quantify the integration layer. For the mean intravenous partition algorithm, the first analysis of the infrared characteristics of a finger vein image, and then validate the model in a variety of traditional segmentation refers to the effect of intravenous extraction, and to explore the feasibility of using the active contour model to partition mean intravenous, and finally maximum curvature segmentation model selected a successful segmentation; for the feature extraction and matching algorithms, validation of seven moments invariants as the reliability characteristics of a finger vein, using the Euclidean distance and the threshold value of classification criteria for the matching and identification; for the score level fusion algorithm, the score of moment matched is normalized by the dual-expectation model, and with the rule of minimizing EER, the averaged increase method and the weighted average method is respectively used to make the best fusion. Then the recognition performance of the single feature law, the averaged increase method and the weighted average method are compared by the three parameters containing EER, FRR, FAR.In this paper, mainly from the following aspects:First of all, This article is researched in on the social background and significance, foreign vein recognition technology refers to the degree of development is introduced, and we also review the research in the field of domestic vein conditions, then point out that the mean vein recognition technology is widely used with promising new research area, lastly the main contents of this paper and structures are given.Secondly, point out two characteristics of the infrared vein images: small contrast of the gray-scale of local area and the existence of polarized light, and the effect of extracting the intravenous is depended on the the ability of segmentation to solve them. In order to assess the performance of the extracting intravenous algorithm under the condition of parameters are controlled, the intravenous images are synthesized. In the finger vein image pre-processing stage, a partial equilibrium is made to increase the global and local image contrast; image range is unified by the Min-Max model normalized. The effect of fixed threshold method, the overall average method, Otsu method and image thresholded method segmenting infrared vein image are verified, in which a fixed threshold method, the overall average and Otsu simply can not afford to split such images, only the image threshold method can overcome the polarization effects to extraction vein, but still can not effectively extract the intravenous in local area of small contrast of the gray-scale.Then, the basic principle and method of active contour model are described in detail, and we system on the evolution process and the advantages and disadvantages of the balloon model, gradient vector flow model, level set methods and geodesic model. And then to explore the possibility of using the C-V model and its improved model to extract the finger vein: Although C-V model can partition the image with no obvious edge, its biggest drawback is that there is no anti-polarizing capacity, which leaded to it have no the basic ability to partition infrared intravenous images; local energy model can not be affected by the polarized light and have ability to extract the weak marginal vein, but it still can not effectively segment the image with small contrast of the gray-scale in some local area.Part IV, maximum curvature model is adopted to extract the vein, then the vein is thinned. Because the curvature map of position-gray profile reflects the location and width of the veins, and in accordance with the actual characteristics of the image, the original algorithm is changed from four direction calculation to five direction, including horizontal direction and with it into 300, 600, 1200, 1500 . Curvature of the profile is calculated to gain the vein. Compared with the image threshold method, local energy C-V model, the greatest curvature model can completely overcome intravenous extraction difficult in some local area of small gray contrast. After the refinement of finger veins, skeleton refers to intravenous is successfully extracted.The fifth part studies extraction and matching of the moment invariant of finger vein skeleton. The use of moment invariants of finger vein skeleton for identification, not only can retain a degree of overall information, but also can avoid the misjudgment by the intravenous changed. First of all, invariance of the seven invariant of the vein image translated, rotated and stretched is verified. And then two finger images are collected twice by the experiment, European distance is used to match feature and the identification is made by the classification criteria of threshold. Experimental results show that the algorithm can efficiently extract moment invariant features, matching and identification can be made by the algorithm.Finally, this dissertation emphasizing the score level fusion algorithm of invariant moment. Basic concepts of the multi-feature fusion algorithm, normalization models and strategies of fusion are introduced . The score of moment matched is normalized by the dual-expectation model, and then with the rule of minimizing EER, the averaged increase method and the weighted average method is respectively used to make the best fusion. Further more the recognition performance of the single feature law, the averaged increase method and the weighted average method are compared by the three parameters containing EER, FRR, FAR. The experimental results show that the performance of the weighted average method in identification is better than he performance of the averaged increase method, and much better than single-feature method'.