Enhanced Multi-View Subspace Clustering via Twist Tensor Nuclear Norm and Constraint Propagation

Multi-view subspace clustering (MVSC) can effectively group multi-view data distributed around several low-dimensional subspaces. Although encouraging results, most existing methods suffer from two typical limitations, resulting in performance degradation. They ignore high-order correlations underlying the data, leading to degeneration of complementary power; addition, they rely on much prior knowledge (e.g., pairwise constraints) for enhancement. In this paper, a novel algorithm called Enhanced Subspace Clustering (EMVSC) is proposed address both limitations. EMVSC exploit and optimally use limited better performance. Specifically, imposes twist tensor nuclear norm representation constructed by stacking view-specific self-representations; exploits constraints whole dataset employing constraint propagation, which propagates constrained samples unconstrained samples. To efficiently optimize EMVSC, an extended intact augmented Lagrangian method derived with good convergence. Experimental results seven standard databases demonstrate its efficacy.