Research On Extended Support Vertor Machines For Human HandBack Vein Recognition

As a new biometric recognition technology, the vein recognition has only been emerged in recent years. It is an effective and security biometric recognition method based on the theory which the human hemoglobin in the blood can absorb a specific wavelength of infrared. However, the core technology of vein recognition is mainly developed in Japan and South Korea, and still at the research stage at home, and there are no commercial products in our country. Because the vein recognition has many advantages such as faster, safer, non-contact, universality, convenience, etc., researching on the vein recognition is very important and it also has a promising market prospects.The thesis studies on the key technologies of vein recognition based on the collection and analysis about the new development of biometric technology at home and abroad in recent years. By a large number of experiments, the main researches are focused on human backhand vein recognition in the thesis as following:1. On the vein image preprocessing stage, firstly, the target vein image is located, making normalization on the target image, filtering and denoising operations which can help to enhanced the clarity and extract the original image for the recognition. Secondly, the Niblack method is used to binarize the original image, smoothing the image again. Thirdly, the massive noise after binarization is filtered, then refining, deburring, and extracting more realistic vein streak line image in order to extract out more detail features for the next step.2. On the feature extraction stage, by refining and burring the vein skeleton, the basic skeleton which is consistent with the original vein after this preprocessing is obtained. This paper mainly uses the method to calculate the seven invariant moment of the image as a data source for the vein recognition.3. On the vein recognition stage, SVM (support vector machine) and its extension algorithms are used as modeling tools. The80%samplings among the seven invariant moments in the vein skeleton will be used as training dataset for the SVM input to establish the model. Then, the other20%samplings will be used as testing dataset to compare it with the venous samplings to determine if the both belong to a same person’s veins.Finally, for each module of the system, The Matlab is used to implement the simulation. In the simulation, a good result is obtained. Also, the best classification results of SVM and BVM (Ball Vector Machine) are compared by using different kernel functions. The results shown that Niblack segmentation method and the polynomial kernel function in SVM can get a better result both in recognition accuracy and time consuming, showing the method usde in this paper is more effective and practical.