Abstract:Latent low-rank representation (LLRR) is a critical method to uncover the low-dimensional subspace structure embedded in high-dimensional data, especially when the data sampling is insufficient and/or extremely corrupted. However, LLRR uses the nuclear norm as a convex relaxation of the rank function, and thus the rank may not be well approximated in practice since the nuclear norm treats all singular values equally. To overcome this deficiency, in this paper, we propose a new model, TS1-LLRR, which extends LLRR along with its corresponding subspace clustering algorithm. We first adopt the non-convex transformed Schatten-1 (TS1) function, a better approximation to the rank of the matrix than the nuclear norm, to model the coefficient matrix. The proposed model works well in learning intrinsic rank information and thus has a stronger ability to capture the subspace structure. Furthermore, to solve the proposed non-convex model, we design an efficient iterative algorithm by using the alternating direction method of multipliers (ADMM) framework. This algorithm performs well in both accuracy and convergence speed, and the resulting subproblems have closed-form solutions. Last, we construct the affinity matrix for spectral clustering by further exploiting the angular information of the principal directions of the low-rank representation, which enhances the capability of investigating the inherent correlation between data points. Extensive experimental results on image clustering and motion segmentation show that our proposed subspace clustering method outperforms the compared methods.

Weighted Schatten p-norm minimization with logarithmic constraint for subspace clustering

On the Schatten P-Quasi-norm Minimization for Low-Rank Matrix Recovery

Subspace Clustering Based on Latent Low Rank Representation with Weighted Nuclear Norm Minimization.

LatLRR for subspace clustering via reweighted Frobenius norm minimization

Subspace Clustering Based on Latent Low-Rank Representation with Transformed Schatten-1 Penalty Function

Recovery of Low-Rank Matrices Based on the Rank Null Space Properties.

Subspace clustering via joint l1,2 and l2,1 norms

Robust Subspace Clustering via Smoothed Rank Approximation

Analyzing the Weighted Nuclear Norm Minimization and Nuclear Norm Minimization based on Group Sparse Representation

Robust Subspace Clustering via Tighter Rank Approximation

Bi-Nuclear Tensor Schatten-p Norm Minimization for Multi-View Subspace Clustering

Improving Generalization of Double Low-Rank Representation Using Schatten-p Norm

Affinity adaptive sparse subspace clustering via constrained Laplacian rank

Data-Dependent Sparsity for Subspace Clustering.

Analyzing the Group Sparsity Based on the Rank Minimization Methods.

Confidence level auto-weighting robust multi-view subspace clustering

Subspace Weighted L(2,1) Minimization For Sparse Signal Recovery

Efficient Low-rank Identification via Accelerated Iteratively Reweighted Nuclear Norm Minimization

Subspace weighted ℓ 2,1 minimization for sparse signal recovery

Rank–sparsity Balanced Representation for Subspace Clustering

Latent Low-Rank Representation With Weighted Distance Penalty for Clustering