Comparing the Performance of the Discriminant Analysis and RBF Neural Network for Face Recognition
Raul Queiroz Feitosa, Carlos Eduardo Thomaz, Álvaro Veiga, Marquês de São Vicente · 1999
Among the many methods proposed in the literature for face recognition, those relying on the so called eigenfaces have been explored with great interest in the last few years. In those methods the face images are initially subjected to a PCA stage (Principal Component Analysis) for dimensionality reduction and then applied to a classifier. This work evaluates and compares two eigenface based face recognition systems, using two different classifiers: a) the LDA (Linear Discriminant Analysis) classifier, and b) a Gaussian Mixture Model RBF (Radial Basis Function) neural network. Extensive experiments using the ORL Face Database indicate that the more general model underlying the RBF classifier does not bring any significant performance improvement compared with the simpler and less computation intensive LDA approach.