TY - JOUR

T1 - Salient object detection employing robust sparse representation and local consistency

AU - Liu, Yi

AU - Zhang, Qiang

AU - Han, Jungong

AU - Wang, Long

N1 - This is the author’s version of a work that was accepted for publication in Image and Vision Computing. Changes resulting from the publishing process, such as peer review, editing, corrections, structural formatting, and other quality control mechanisms may not be reflected in this document. Changes may have been made to this work since it was submitted for publication. A definitive version was subsequently published in Image and Vision Computing, 69, 2018 DOI: 10.1016/j.imavis.2017.10.002

PY - 2018/1

Y1 - 2018/1

N2 - Many sparse representation (SR) based salient object detection methods have been presented in the past few years. Given a background dictionary, these methods usually detect the saliency by measuring the reconstruction errors, leading to the failure for those images with complex structures. In this paper, we propose to replace the traditional SR model with a robust sparse representation (RSR) model, for salient object detection, which replaces the least squared errors by the sparse errors. Such a change dramatically improves the robustness of the saliency detection in the existence of non-Gaussian noise, which is the case in most practical applications. By virtual of RSR, salient objects can equivalently be viewed as the sparse but strong “outliers” within an image so that the salient object detection problem can be reformulated to a sparsity pursuit one. Moreover, we jointly utilize the representation coefficients and the reconstruction errors to construct the saliency measure in the proposed method. Finally, we integrate a local consistency prior among spatially adjacent regions into the RSR model in order to uniformly highlight the whole salient object. Experimental results demonstrate that the proposed method significantly outperforms the traditional SR based methods and is competitive with some current state-of-the-art methods, especially for those images with complex structures.

AB - Many sparse representation (SR) based salient object detection methods have been presented in the past few years. Given a background dictionary, these methods usually detect the saliency by measuring the reconstruction errors, leading to the failure for those images with complex structures. In this paper, we propose to replace the traditional SR model with a robust sparse representation (RSR) model, for salient object detection, which replaces the least squared errors by the sparse errors. Such a change dramatically improves the robustness of the saliency detection in the existence of non-Gaussian noise, which is the case in most practical applications. By virtual of RSR, salient objects can equivalently be viewed as the sparse but strong “outliers” within an image so that the salient object detection problem can be reformulated to a sparsity pursuit one. Moreover, we jointly utilize the representation coefficients and the reconstruction errors to construct the saliency measure in the proposed method. Finally, we integrate a local consistency prior among spatially adjacent regions into the RSR model in order to uniformly highlight the whole salient object. Experimental results demonstrate that the proposed method significantly outperforms the traditional SR based methods and is competitive with some current state-of-the-art methods, especially for those images with complex structures.

KW - Salient object detection

KW - Robust sparse representation

KW - Local consistency

KW - Complex structures

U2 - 10.1016/j.imavis.2017.10.002

DO - 10.1016/j.imavis.2017.10.002

M3 - Journal article

VL - 69

SP - 155

EP - 167

JO - Image and Vision Computing

JF - Image and Vision Computing

SN - 0262-8856

ER -