Two-fold regularization for kernel Fisher discriminant analysis in face recognition
Sang-Ki Kim, Kar‐Ann Toh, Sangyoun Lee · IEICE Electronics Express · 2009
Due to the inherent nature of kernel implementation, the kernel Fisher discriminant suffers from the small sample size problem. In this paper, we introduce a novel variant of the kernel Fisher discriminant formulation to circumvent this problem. By adopting a two-fold regularization scheme on the scatter matrices, we show both effectiveness and reliability of the proposed method particularly regarding the small sample size and the lack of dimensionality issues.