Support Vector Machine And Its Application In Biometric Recognition

The status of network information security has become increasingly prominent with the popularization and vigorous development of Internet technology,and the network attacks launched against network information are becoming more and more intensive.This has brought new challenges to the major portals that guarantee network information security.Using human-machine verification technology to resist large-scale network attacks,the main solution is verification code technology.As verification code technology is more and more widely used on major websites and APPs,verification code cracking technology is becoming more mature and cheaper.In order to prevent network attackers who have cracked verification code technology,the security of network information is further strengthened.More and more researchers began to explore new types of human-machine verification technology.In this context,this thesis mainly studies the biometrics generated during the operation of the mouse to draw the specified trajectory,and constructs a human-machine recognition model for this biometric to predict whether the person or computer program submitted the verification trajectory.Different from the traditional simple judgment that only relies on the user's pattern recognition ability,the behavior verification code that adds biometric recognition uses double authentication: 1)Compare the similarity of the user feedback information with the hidden information in the verification picture,the purpose is to judge whether the user is Recognize the information shown in the verification diagram;2)According to the process of the mouse returning the verification information in the form of drawing the track,the biological characteristics generated can effectively distinguish the person or the computer program.After extracting the user's biometric information,how to effectively perform human-machine recognition is another research focus of this thesis.After analyzing the advantages and disadvantages of various machine learning algorithms,this thesis intends to use support vector machine as a classification model.After researching the basic theory of support vector machines,this thesis attempts to propose a new support vector machine acceleration algorithm for the inevitable large-scale data set problem in the process of using support vector machines to build biometric models.Inspired by the idea of acceleration and the K-Means algorithm,it is proposed to use the K-Means algorithm to reduce the training sample set by a certain percentage before training the support vector machine model,so as to achieve the purpose of increasing the convergence speed of the support vector machine model.In order to verify the feasibility of the improved algorithm,verification experiments were carried out on several commonly used artificial data sets in the field of machine learning.The experimental results show that the training sample set reduced by the K-Means algorithm maintains the classification accuracy of the support vector machine model.On the premise of a small decrease,only using about 10% of the sample data of the entire training sample set can achieve a classification accuracy rate similar to the support vector machine trained with the entire training sample set,which can greatly improve the training of the support vector machine model.The efficiency and convergence speed prove the feasibility of this improved scheme.Finally,this accelerated support vector machine algorithm is applied to the identification of biological features.The experimental results show that the improved support vector machine model can achieve the purpose of rapid and effective human-machine classification,and provides a new guarantee for network information security.