Structure-Constrained Low-Rank and Partial Sparse Representation with Sample Selection for image classification. (November 2016)

• Record Type: Journal Article
• Authors: Liu, Yang; Li, Xueming; Liu, Chenyu; Liu, Haixu
• Is Part Of: Pattern recognition. Volume 59(2016:Nov.)
• Page Start: 5
• Page End: 13
• Publication Date: 2016-11
• Subjects: Sparse coding -- Low-rank -- Dictionary learning -- Image classification -- Structured sparsity; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.01.026
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: In this paper, we propose a novel Structure-Constrained Low-Rank and Partial Sparse Representation algorithm for image classification. First, a Structure-Constrained Low-Rank Dictionary Learning (SCLRDL) algorithm is proposed, which imposes both structure and low-rank restriction on the coefficient matrix. Second, under the assumption that the coefficient of test sample is sparse and correlated with the learned representation of training samples, we propose a Low-Rank and Partial Sparse Representation (LRPSR) algorithm which concatenates training samples and test sample to form a data matrix and finds a low-rank and sparse representation of the data matrix over learned dictionary by low-rank matrix recovery technique. Finally, we design a Sample Selection (SS) procedure to accelerate LRPSR. Experimental results on Caltech 101 and Caltech 256 show that our method outperforms most sparse or low-rank based image classification algorithm proposed recently. Abstract : Highlights: We propose a Structure-Constrained Low-Rank Dictionary Learning algorithm and develop its optimization strategy. We propose a Low-Rank and Partial Sparse Representation algorithm and develop its optimization strategy. We prove that the solution to LRPSR is block sparse for independent subdictionaries. We design a Sample Selection procudure to accelerate LRPSR. Experimental results show that our proposed method outperforms most sparse or low-rank based image classification algorithms proposed recently.