Fast kernel Fisher discriminant analysis by approximating principle component analysis

The feature extraction efficiency of KFDA (kernel Fisher discriminant analysis) is enslaved to the size of the training sample set. In order for an efficient nonlinear feature extraction, we propose a fast KFDA (FKFDA) in this paper. This FKFDA consists of two parts. First, we select a portion of training samples based on two criteria produced by approximating principle component analysis in the kernel space. Then, referring to the selected training samples as nodes, we formulate FKFDA to improve the efficiency of feature extraction by assuming that nodes can replace all the training samples to express the discriminant vectors. The extraction of a feature from a sample in FKFDA requires only calculates as many kernel functions as the nodes. Moreover, the proposed FKFDA can avoid the small sample size problem. Experimental results on face recognition suggest that the proposed FKFDA can generate well classified features.