Multi-view spectral clustering via partial sum minimisation of singular values

This Letter proposes a robust multi-view spectral clustering approach. It first calculates a normalised graph Laplacian for each single view, and then uses them to recover a shared low-rank Laplacian by the low rank and sparse matrix decomposition. To achieve matrix decomposition, partial sum minimisation of singular values is leveraged to design a novel objective function, which can be optimised by the augmented Lagrangian multiplier algorithm to recover a common normalised graph Laplacian. Accordingly, multi-view clustering results can be obtained by taking spectral clustering on the common Laplacian. Experimental results illustrate its effectiveness over other related approaches.