The Study Of Hand Vein Pattern Recognition Method

With the development of the information security requirement of modern society, using biometric character to identify one's identification quickly and exactly thrives. Vein Pattern Recognition is a new contactless biometric technology using IR. It not only offers high accuracy personal identification, but also offers high safety, usability, and can be implemented easily. So it becomes a hot spot stage by stage. This paper has studied it on three points: vein segmentation, skeleton features extraction and vein classification. The researches in this thesis can be summarized as follows:(1)Status quo on Vein Pattern Recognition is summed up. It's principium and traits are introduced firstly, and then the leading algorithms home and overseas are presented following the processing flow. Furthermore, their merits and disadvantages are reviewed respectively. Also, suggestions to future research are proposed.(2)Morphological Image Processing, Image Moments and Support Vector Machine are reviewed. Opening, closing, geometrical invariable moments and Support Vector Machine are chiefly explained.(3)A new method combing Gaussian Lowpass Fliter, Row Median Filter and Column Median Filter to reduce noise is proposed. Then the improved NiBlack algorithm is adopted to segment vein. Opening and closing are used to smooth the vein edge. The effects of the parameters during the segmentation are discussed. A concrete pruning algorithm based-on endpoint, intersection and the length of the burr is also presented.(4)The seven invariable moments proposed by Hu are invariant to translation, rotation and scale change. So they are modified and standardized to describe vein skeletons, and used as the input vector of the classifier.(5)Support Vector Machine is employed to classify vein patterns. The C-SVC from the software kit LIBSVM is used to conduct experiment on a database containing 500 samples. This multi-class classifier adopts One-against-One mode and uses Radial Basis Function as Kernel Function. The result shows that the identification rate is high.