Fuzzy Support Vector Machine And Its Application To Face Recognition

Support Vector Machine (SVM) as the statistical learning theory of VC dimension theoryand structural risk minimization of criteria specific implementation tools was proposed byVapnik and his partners. Compared with traditional neural network learning methods, SVMhas many merits such as global optimum, simple structure, generalization performance and soon. Support Vector Machine is a new and very powerful machine learning method based onstatistical learning theory,especially for the small sample size problem, chose the optimalseparating hyperplane as the discriminant function, at the same time interval to maximize theclassification as a condition,then the classifiedtranslate will be directly into a quadraticprogramming problem. By introduced the kernel function make the linear inseparableproblem mapped into a high dimensional space into linear can be divided skillfully. Due touse nuclear mechanism, so the computational complexity of the problem was no increase.Support vector machine reflected many of its unique advantages to solve the problem of smallsample, has become the preferred field of pattern recognition in the current internationalclassifier.The face recognition technology is an important biometric authentication technology, hotand difficult of the field of pattern recognition and image processing. It used computer toanalysis the face image, and extract the effective identification information, compared withthe known face in the information database information so as to achieve the identificationidentify identity. After years of research, has been made many important results. However, thehuman face as a non-rigid body has a large deformation, many factors, susceptible tointerference and other characteristics. It is a high dimension, nonlinear, small sample sizeproblem relatived to the image vector dimension. Some of the traditional pattern recognitionmethods were prone to learning or owe learning phenomenon. This article based on fuzzysupport vector machine (FSVM) theory to open analysis, obtain a new fuzzy support vectormachine by explored the construction of fuzzy membership function, the selection of thekernel function and clustering rules. Ultimately, the new design of the fuzzy support vectormachine was applied to face recognition technology.In this thesis, the main work and includes were the following aspects: (1) Described the theoretical basis of support vector machine in detail, analyzed thelimitations of the empirical risk minimization and the superiority of the structural riskminimization. In-depth study of some of the main support vector machine learning algorithm.Put forward a Fuzzy Support Vector Machine (DM-FSVM) based on the density of dualmembership.(2) Designed a screening sample of rules to improve the K-means clustering algorithm.According to the simple rules of clustering to carry out simple the original data set sample andensured the accuracy of the premise. To improve the capacity and speed of the algorithm tohandle large-scale sample, and analyzed the improved algorithm from multiple perspectives.(3) Introduced human face recognition method, summarized the mainstream algorithm.On the basis of the basic principles of PCA and ICA in detail, gived a two-way PCArecognition algorithm based on local image, and used to implement the facial featureextraction.(4) Used the fuzzy support vector machine as classifier which put forward in this paper,verify the effectiveness of the algorithm by experiments on the ORL face database and ARface database.