Semi-Supervised Classification based on Local Sparse Representation

The performance of graph-based semi-supervised classification depends on the structure of graph. However, how to construct a graph that correctly reflects the sample distribution is a non-trivial job, because of the noisy samples and features, especially for high-dimensional data. Sparse representation has been used to construct graphs for graph-based semi-supervised learning, which is shown to be robust to noisy samples and features, but it often asks for huge computational resources. In this paper, to address these problems associated with applying sparse representation for graph-based semi-supervised classification, we propose a method coined as Semi-Supervised Classification based on Local Sparse Representation (SSC-LSR in short). SSC-LSR first utilizes k nearest neighbors of a sample to compute the reconstruction weights, instead of the whole samples. Then, these weights are adopted to define a graph. Next, it incorporates this graph into the widely used Gaussian Random Filed and Harmonic Functions to classify unlabeled samples. Experimental results on face datasets demonstrate that the proposed method not only has higher classification accuracy than traditional methods, but also is robust to the input parameters. Copyright © 2015 Binary Information Press.