Font Size: a A A

Face Recognition Based On Support Vector Machine Technology

Posted on:2009-07-22Degree:MasterType:Thesis
Country:ChinaCandidate:P DuFull Text:PDF
GTID:2208360272956127Subject:Computer technology
Abstract/Summary:PDF Full Text Request
Face recognition is an active research field in identification technology of living things. At Present, face recognition continues to be a hot topic in pattern recognition field due to its wide range of applications such as distinguishing somebody from others and controlling authority. Support vector machines, the implementation of Structural Risk Minimization rules, have some attractive merits, such as global optimization, simple structure, generalize abilities and so on. So the development and application about SVM developed widely in recent years. Human faces have large variation in shape at different time, compared with the dimension of the face image vectors, face recognition is also a high dimensional and nonlinear small-sample problem. This thesis gives a improved algorithm to SVM and Feature Extraction, respectively, makes it more adapt to face recognition problems.The main work reads as follows:(1) We have researched the methods of pretreatment and feature extraction for face image, used the illumination compensation and the histogram equalization and so on. Before we adopt the Linear Discriminant Analysis (LDA) method for feature extract, the Principal Components Analysis (PCA) method is used for dimension reduction, then, feature vectors are gutted. The implement algorithm is also available.(2) There are two conversion methods, decompose one multi-classification problem to many two-classification or one two-classification. A two-stage method is provided, firstly, clustering all face images to k small sample sets by comparability; secondly, adopting one-against-restl SVM classification method on sample sets.(3) Finally experiments are carried out on YALE database and ORL human face database to verify validity of our algorithm. We compare these multi-classification methods in the aspects of training time, testing time and recognition rate. The conclusion is our method is better fit for human face recognition than any others.
Keywords/Search Tags:Support Vector Machine, Face Recognition, Multi-classification Problems, Feature Extraction, PCA, LDA
PDF Full Text Request
Related items