Multi-view subspace clustering with inter-cluster consistency and intra-cluster diversity among views

Multi-view subspace clustering aims to classify a collection of multi-view data drawn from a union of subspaces into their corresponding subspaces. Though existing methods generally make promising performance, fully making use of the diversity and consistency of multiple information leaves space for further improvement of the clustering results. In this paper, we explore two new constraints: inter-cluster consistency among views (ICAV) and intra-cluster diversity among views (IDAV). Based on IDAV, we propose a new regularization term which couples the intra-cluster self-representation matrix and the label indicator matrix. This new regularization term tends to enforce the self-representation coefficients from the same subspace of different views highly uncorrelated. A technique similar to Exclusivity-Consistency Regularized Multi-view Subspace Clustering (ECMSC) is also used to enforce ICAV of self-representation coefficients. Further, we formulate them into a unified model and call it Multi-view Subspace Clustering with Inter-cluster Consistency and Intra-cluster Diversity among views (MSC-ICID). Based on the alternating minimization method, an efficient algorithm is proposed to solve the new model. We evaluate our method using several metrics and compare it with several state-of-the-art methods on some commonly used datasets. The results demonstrate that our method outperforms the state-of-the-art methods in the vast majority of metrics.