ENEE633 Project Report SVM Implementation for Face Recognition

Support vector machine(SVM) is a very popular way to do pattern classification. This paper describes how to implement an support vector machine for face recognition with linear, polynomial and rbf kernel. It also implements principal component analysis and Fisher linear discriminant analysis for dimensionaly reduction before the classification. It implements svm classifier in MATLAB based on libsvm interface as well as scaler and parameter selector, which uses cross validation to find the optimal parameters for each kernel of classifiers. To apply svm on multiclass problem, it applies the one-against-one method due to its simplicity and good performance. It measures and compares different classifiers performance in terms of accuracy and gives the best parameters with the cross validation accuracy for each kernel under different scenarios including different selected training samples such as expressions and illuminations variations, and the amount of training data.