Research On Classification And Matching For Hand Vein Image

Along with the development of information technology, biological identity authentication technology is being received more and more people's attention. As a new method of biological feature recognition, vein recognition, although with a late start, has become the mainstream biometric identification method after several years of development due to its unique advantages.Any kind of identity authentication technology should pass through the process from research to practice. For vein recognition, all kinds of circumstances appeared or may appear must be considered in actual use process. In order to guarantee the real-time and accuracy in vein recognition, this thesis will primarily take vein classification and matching to research.So far, no literatures of vein classification have been found. The vein wavelet moment is chosen after research experiment, which will be used in latter vein classification and matching process. Based on the analysis and research of the other digital image classification technology, a new vein classification method that can be scale applicated is put forward. Firstly, BP neural network classifier, which is widely used in many classification fields, is used to design the vein classifier. And this classifier can get good effect after improved by LM algorithm. In order to judge the rationality of the classification results, a standard that judge the reasonability of classification is put forward in this paper. Considering the BP neural network classifier is a 'supervision' classifier and it will take a lot of time to gather the class number of veins, a new vein classification strategy based on 'clustering and classification' is put forward. This strategy can realize the 'non-supervised' classification of vein images.The ant colony optimization is used to get every vein's class number which will be the target output of the BP neural network through achieving the clustering part of the strategy. In addition,condering the BP neural network's generalization ability is insufficient and can easily sink into local optimum, a research on SVM vein classifier based on ant colony optimization is put forward.The experiment results prove that the design of this classifier is effective.For vein matching, this paper has a research on the mechods based on gray and characteristic.After experiment, the nearest neighbor method vein matching method that based on vein wavelet moment is adopted. This method has a high recognition rate from the experiment results. It also proves that the vein feature selection,vein feature classification and vein matching of the whole paper are valid.Finally, achievement and deficiency of thesis are summarized, ideas on improving vein classification and matching methods are introduced, and vein recognition research forecast is also proposed.