Face recognition based on multi-class SVM

Abstract

Support vector machine (SVM) provides high performance in generalization, processing small samples, and tackling high-dimensional data. Based on the advantages of SVM, an approach is proposed in this paper, adopting multi-class SVM to realize face recognition. In the approach, principle component analysis (PCA) is used firstly to reduce dimensions so that feature extraction is carried out on face images. Then a method based on one-versus-all svm is implemented to realize multi-class classification on feature vectors of the face images. Results of experiments applied to ORL and Yale face databases show that our approach is effective. By the one-versus-all SVM method, we can respectively obtain recognition rates as high as 93.5% in ORL face database, and 97.3% in Yale face database.