The Study Of Finger Vein Recognition Methods

Along with the arriving times of information, people pay high attention to information security and secrecy. Fingerprint recognition is considered as a more comparatively mature technology than other biological feature recognition such as iris recognition and face recognition, which has well applied in many aspects of our life such as safety fixture, entrance guard, examining system, and so on. However, fingerprint recognition has many unavoidable defects, such as over-wet, over-drying, cruel skin or contamination of finger that would affect the recognition results. However, the finger vein recognition technology can completely overcome the defects listed above, making the identity authentication easier and faster.The dissertation discussed the technology of finger vein recognition, and it is consisted of 5 parts below:First, it discussed the background, purpose and the meaning of the technology of finger vein recognition, which has many merits such as living body recognition, interior characteristic, non-contacting and high safety class, and so on. Therefore the research is meaningful and feasible. By means of analyzing the evolution of the technology home and abroad we got the clear goal of the research. At the same time, this part gives the stressing and difficulty points of the research and the overall framework of the system.Second, starting from the basic imaging principle of finger vein, comparing auxiliary devices in aspects such as different optical devices, imaging devices and light filters, we selected devices that meet our requirements, and made a collection device. Computer tests indicated that images collected by the device meet the experiment requirements.Third, we preprocess the collected image to make it ready for the future feature extraction. We make the image to a gray one to reduce the scale of data processing. In order to avoid the effect of area on and out of the finger profile, we adopt a method which implement by iterative threshold to extract the finger area and label the profile. We also use the combined filter on the image to make the model more precision before the image segmentation. After image segmentation, there are many dots and block noises on the image. So we can compute the area of the connected domain to wipe off the area which less then the given threshold. The end of this part, we propose a method which normalize the image width.In the fourth part, we implement four recognition methods, that is, based on templated matching, based on moment character, based on wavelet moment and based on wavelet moment fusing with PCA and LDA transform. Templated matching is a method which is often used in pattern recognition. In the thesis, we firstly implement a algorithm to smooth the template and then made the template matching. The experiment demonstrated that after smooth processing, template matching has a high recognition rate but low in speed. Aimed at the ocassion that different finger samples has different length, a method of extracting the sub-image blocks of the same size at the certain intervals is given. The last three methods of recognition extract characteristics based on the sub-image block extracting. The fourth one, wavelet moment fusing with PCA and LDA transform performs very well in both speed and recognition rate.Finally, it analyzed the experiment results of the methods mentioned above, and validated the algorithms based on template matching, character moment, wavelet moment and wavelet moment fusing with PCA and LDA transform with library of finger vein separately, and conducted corresponding analyses to recognition results.