Towards robust subspace recovery via sparsity-constrained latent low-rank representation

摘要: We present a Sparse Latent Low-rank representation approach for robust visual recovery.This constructs the dictionary using both observed and hidden data.A low-rank with enhanced sparsity can be derived.Extensive experiments have confirmed superiority of proposed method. Robust recovery subspace structures from noisy data has received much attention in analysis recently. To achieve this goal, previous works developed number based methods, among which Low-Rank Representation (LRR) is typical one. As refined variant, LRR to relieve insufficient sampling problem. However, they fail consider observation that each point represented by only small subset atoms dictionary. Motivated this, we (SLL) method, explicitly imposes constraint on encourage sparse representation. In way, selecting few points same subspace. Its objective function solved linearized Augmented Lagrangian Multiplier Favorable experimental results clustering, salient feature extraction outlier detection verified promising performances our