Consistent multi-view subspace clustering with local structure information

Multi-view subspace clustering has attracted extensive attention in recent years because it can fully utilize the inherent characteristics of each view data. The low-rank structure in many current successful multi-view subspace clustering methods to describe the consistency is based on the minimization of the nuclear norm, but it is easily dominated by large eigenvalues, thereby resulting in an inaccurate low-rank structure. Besides, local structural information, which can reduce the distance between similar points, has not been fully employed in multi-view subspace clustering algorithms. In this article, an improved consistent multi-view subspace clustering model with local structure information (CLSI-MSC) is proposed to overcome the aforementioned limitations. Firstly, the Schatten-p norm is substituted for the nuclear norm, thus ensuring that the low-rank structure of the consistency component can mine the consistency information better. Additionally, the group effect to preserve the local structural information between data points is introduced into the model to reduce the distance between similar points. Finally, \(l_{2,1}\) regularization is introduced to make the model more robust against noise and outliers. The experimental results demonstrate that the proposed method is superior or comparable to other comparison methods on five benchmark datasets.