Kernel Non-Locality Preserving Projection and Its Application to Face Recognition

Abstract

Non-locality preserving projection (NLPP) is a kind of feature extraction technique based on the characterization of the non-local scatter. Due to NLPP is a linear algorithm in nature, it cannot address nonlinear problem in recognition, so a novel subspace method, called Kernel Non-locality Preserving Projection (KNLPP) discriminant analysis, is proposed for face recognition. Experimental results on two popular benchmark databases, FERET and Yale, demonstrate the effectiveness of the proposed method.