Refined Kernel Principal Component Analysis Based Feature Extraction

Kernel principal component analysis (KPCA) has been widely applied in pattern recognition areas, but it endures the high store space and time consuming problems on feature extraction in the practical applications. In this paper, we propose a novel Refined kernel principal component analysis (RKPCA) based feature extraction with adaptively choosing the few samples from the training sample set but with less influence on recognition performance in the practical applications. Experimental results on seven datasets show the proposed algorithm achieves the approximate error rates but only about 20%-30% training samples. RKPCA performs well on the conditions of high computation efficiency but not a strict on recognition accuracy.