Adaptive Tensor Rank Approximation for Multi-View Subspace Clustering

Multi-view subspace clustering under a tensor framework remains challenging problem, which can be potentially applied to image classification, impainting, denoising, etc. There are some existing tensor-based multi-view models mainly making use of the consistency in different views through nuclear norm (TNN). The diversity means intrinsic difference individual view is always ignored. In this paper, new tensorial model proposed, jointly exploits both and each view. representation decomposed into view-consistent part (low-rank part) view-specific (diverse part). A adaptive log-determinant regularization (TALR) imposed on low-rank better relax multi-rank, sparsity diverse ensure connectedness property. Although TALR minimization not convex, it has closed-form analytical solution its convergency validated mathematically. Extensive evaluations six widely used datasets executed our demonstrated have superior performance.