Research On Key Technologies Of Finger-vein Recognition

Biometric identification technology is booming in recently years, and now, many different kinds of biometric identification methods have been launched on the market, such as finger-print recognition, iris recognition and face recognition, to name a few. But we have to admit that, all the methods put us in a dilemma to decide which one is better considering the easiness and reliability. Different from the traditional biometric identification methods, finger-vein recognition shows so many characteristics, for example, imperceptibility, hard to imitate, easy to use, and high precision, that it has been taken as the second generation of biometric identification technology. Although this technology reveals many advantages, there are still some pending problems about imaging, image preprocess, feature extraction, and image matching, and the objective of this study is to solve these problems mentioned above.This dissertation will lay particular emphasis on following aspects.In view of the problem of imaging system development that is there is no general designing method or designing framework, considering that the imaging system is the most important part of the entire recognition system, one simple and practical imaging system is developed by the synthesis of existing designing methods, and the detailed designing process is described in this dissertation. The acquired images using this system are validated by experiments.In view of the problem of image preprocess, based on the research on the gray level normalization, image zooming, image enhancement, a new method is proposed which combines the traditional Sobel operator with Canny operator about the finger positioning step. Combined with the dual-threshold method, this new method can remove the coarse edges from Sobel operator and lots of noise from Canny operator.In view of the problem of feature extraction, by the synthesis of a lot of existing feature extraction methods, one more simple and practical method has been applied, which mainly consists of the four steps of orientation-valley detection, fuzzy enhancement, dynamic threshold segmentation, and imaging thinning. Every single step will be elaborated in this dissertation, several different threshold segmentation methods will be compared so as to choose the right segmentation method. In view of the problem of image matching, based on the research on the traditional template matching method, a kind of modified template matching method is applied, and considering the orientation features of finger-vein, two methods based on Gabor filters are also applied. In consideration of the big impact of image rotation on the matching methods, as the result of research on the existing finger-vein anti-rotation methods, two new methods based on Hough transformation are proposed. Although these two methods are some kind of approximation processes, they can also enhance the algorithms'ability to be adaptive to image rotation markedly.