Kernel methods have been applied to many linear feature analysis classifiers to generate nonlinear classifiers for improved classification performance. The recently proposed kernel class-dependence feature analysis (KCFA) method extends linear correlation filter technology to kernel correlation filters, greatly improving the classification performance. In this paper, we compare the KCFA method with the kernel discriminant analysis (KDA) method and show that the KCFA and the KDA result in the same representation subspace and the relationship between them is similar to the relationship between the orthogonal and the simplex signal representations in digital communications. We present face recognition results to illustrate that the KCFA method is preferable to the KDA method.
展开▼