Multi-view subspace clustering with intactness-aware similarity

Abstract

Multi-view subspace clustering, which aims to partition a set of multi-source data into their underlying groups, has recently attracted intensive attention from the communities of pattern recognition and data mining. This paper proposes a novel multi-view subspace clustering model that attempts to form an informative intactness-aware similarity based on the intact space learning technique. More specifically, we learn an intact space by integrating encoded complementary information. An informative similarity matrix is simultaneously constructed, which enforces the constructed similarity to have maximum dependence with its latent intact points by adopting the Hilbert–Schmidt Independence Criterion (HSIC). A new explanation on the advantages of such intactness-aware similarity has been provided (i.e., the similarity is learned according to the local connectivity). To effectively and efficiently seek the optimal solution of the associated problem, a new ADMM based algorithm is designed. Moreover, to show the merit of the proposed joint optimization, we also conduct the clustering in two separated steps. Extensive experimental results on six benchmark datasets are provided to reveal the effectiveness of the proposed algorithm and its superior performance over other state-of-the-art alternatives.