Symbolic Kernel Fisher discriminant method with a new RBF Kernel function for face recognition

Abstract

In this paper, we present a new radial basis kernel function (RBF) in symbolic kernel Fisher discriminant analysis (symbolic KFD) to extract nonlinear interval type features for face recognition. The kernel-based methods form a powerful paradigm, they are not favorable to deal with the challenge of large datasets of faces. We propose to scale up training task based on the interval data concept. Our investigation aims at extending KFD to interval data using new RBF kernel function. We adapt symbolic KFD to extract interval type nonlinear discriminating features, which are robust enough to varying facial expression, viewpoint and illumination. In the classification phase, we employ the minimum distance classifier with the squared Euclidean distance measure. The new algorithm has been successfully tested using four databases, namely, the ORL face database, the Yale face database, the Yale face database B and the FERET face database. The experimental results show that the symbolic KFD with the new RBF kernel function yields improved performance.