Font Size: a A A

Research And Application Of Multi-Class Classfiction On Support Vector Machine

Posted on:2011-07-15Degree:MasterType:Thesis
Country:ChinaCandidate:Y N LangFull Text:PDF
GTID:2218330338967143Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Support Vector Machine(SVM) is a new kind of machine learning method based on statistical learning theory that put forward by Vapnik etc. SVM approach based on structural risk minimization principle has good generalization ability and accuracy with considering the empiric risk and incredible risk. It integrates the optimal hyper-plane, kernel function, convex quadratic programming technology, etc. It can solve some problems effectively like "over fitting", "under fitting", "dimension disaster" and "local minimum point". SVM has been the focus of machine learning research, and it is widely applied in many fields such as pattern recognition, regression estimation. Due to the SVM method was put forward for the binary classification problem at first, how to expand binary classification to multi-class classification is classified as an important problem of support vector machine research. In this paper, the main work is as follows:1. Make an introduction on machine learning, statistical learning theory and development and research situation of support vector machine. The theory and the algorithm of support vector machine are introduced in detail. The kernel function theory and the parameter selection are discussed.2. Relative distance is a yardstick to measure inter-class difference, which is proposed by analyzing common multi-class classification methods. New binary trees are established by considering the distance and distribution of classes. The training time and efficiency of improved methods are proved by results of experiment.3. New binary trees based on SVM are applied to facial orientation recognition. First, preprocesses the images. Then, extracts the feature of eyes. At last, it achieves good recognition effect to classify the eigenvalue with new classification methods of binary tree based on SVM.
Keywords/Search Tags:Statistical Learning, Support Vector Machine, Multi-class Classification, Facial orientation recognition
PDF Full Text Request
Related items