Subspace Clustering via Sparse Graph Regularization

Abstract

Subspace clustering aims to segment data drawn from a union of linear subspaces. Recently various self-representation based methods have been proposed and achieve much more successful performance. Smooth Representation clustering (SMR) is one of these methods, which does self-representation coding with a graph regularization term and enjoys the grouping effect. In this paper, we propose a new subspace clustering method via sparse graph regularization, modifying the traditional graph regularization term of SMR into a new sparse graph regularization term, which is more robust against noise and outlying data. We theoretically study the nice properties of the proposed method and provide an efficient algorithm to solve the new spare graph regularized subspace clustering problem. Experiments on several subspace clustering tasks show that our method gets significantly better performance than the state-of-the-art methods.