Reweighting Recognition Using Modified Kernel Principal Component Analysis via Manifold Learning

Abstract

Kernel Principal Component Analysis (KPCA) is a widely used technique in the dimension reduction, de-noising and discovering nonlinear intrinsic dimensions of data set. In this paper we describe a reweighing kernel-based classification method for improving recognition problem. Firstly, we map the training samples to the feature space by non-linear transformation, and then perform principal component analysis(PCA) using the selected kernel function in the feature space, and get the linear representation of testing samples in the feature space. Secondly, by using the idea of reweighting, we select the similarity between testing sample and each training sample as the weight of reweighting, then take the final weight as the criteria of classification. The experimental results demonstrate that our method is more accurate than Support Vector Machine (SVM) classification method and Linear Discriminant Analysis (LDA) classification. In addition, the number of training samples that our method need is much smaller than some other methods.