一种基于支持向量机的多生物特征融合身份识别方法分析-analysis of multi-biometric feature fusion identification method based on support vector machine.docxVIP

Research on Multi-Biometrical Identity RecognitionMethod Based on SVMAbstractIn today’s society which information has highly developed, biometric technology provide an effective way for identification and subject to the attention of more and more researchers. Compared to a single feature recognition, multi-biometrical identity recognition has the following characteristics: 1)greater reliability which can improve the performace and fault tolerance of the system and reduce noise; 2)broader applicability which make up the single-frature’s shortcomings that it may be not suitable for some people; 3)stronger security and so on. It has a very broad application prospects.The Support Vector Machine is a new tool for machine learning based on optimization methods of staristical learning theory. It has the ability to solve high dimension, nonlinear and small samples. It has been widely used in the field of bioinformatics, artificial intelligence and so on. Using the SVM for biometric identification has a very important significance.The main work of this paper is:Because of the insufficient of the primitive face images, fingerprint images and iris images, these images have to be processed to eliminate noise and extract features. Use SVM to build the classifier and to train and test the features. The experments has feasibility and a high recognition rate.An improved Gaussian function is researed to build the kernel function. Select the appropriate parameters to improve the learning and classification abilities of the Gaussian kernel function. Its legitimacy has been proved. The improved Gaussian kernel is uesd for biometric experiment. Compared with the RBF kernel function, it improved the performace of the system.A decision level fusion method of the three single feature recognition is studied. Pairwise fusion and fusion of the three are separate experiments. The three biometric databases are simulately correspond to establish a multi-feature database. The experiments suggest