Two-phase Framework for KFD with Application to Face Recognition
Jiande Wu, Yugang Fan · 2008
After analyzing the kernel Fisher discriminant analysis (KFD), a simple two-phase framework for KFD is proposed in the paper. The base idea is that the nonlinear mapping function, which is used to map the input data space into feature space F, is approximated via Nystrom method. Then, the approximate feature of the input data is used in the Fisher linear discriminant analysis (LDA). Following this framework, the paper presents a modified KFD based on approximate nonlinear mapping (ANM-RLDA) for face recognition. ANM-RLDA is nonlinear extension of the regularized LDA (R-LDA). The proposed algorithm is tested and evaluated using the UMIST face database. The experimental results show ANM-RLDA good performance.