A novel improved sampling algorithm
Zhiying Tan, Yong Feng · 2011
In the kernel principal component analysis (KPCA) and manifold learning to reduce the dimensionality, the training set plays a very important role. In this paper, we develop a novel method to select the training samples to reduce the computation of feature extraction and to improve the accuracy of image restoration. The developed method consider the density distribution of the samples and retained the border samples. Experiments on several data sets illustrate that the feature extraction derived from the selected training set is much more efficient than from the original sample set with KPCA. And the novel sampling method can maintain the intrinsic dimension of manifolds formed by the sample points.