Neighbor-relation aware low-rank multi-view subspace clustering

Low-rank multi-view subspace clustering seeks to uncover complex relationships in high-dimensional data by leveraging multiple views and low-rank representations. However, two key challenges persist: (1) Existing methods emphasize low-rank consistency through nuclear norm regularization but often overlook the unique low-rank features of individual views. (2) Constructing the affinity matrix directly from the coefficient matrix can introduce excessive connections, degrading clustering performance. To address these challenges, we propose neighbor-relation aware low-rank multi-view subspace clustering (NLMSC), a unified model that integrates view-specific low-rank structure learning with neighbor-aware affinity fusion. NLMSC decomposes the coefficient matrix into smaller factor matrices to capture the underlying low-rank structures of each view and constructs a sparser, more interpretable affinity matrix by reliable neighbor selection. These affinity matrices are then fused into a block-diagonal consensus graph through a self-weighting strategy for spectral clustering. Extensive experiments on diverse multi-view datasets demonstrate NLMSC's superior performance and effectiveness in challenging clustering tasks.