Sparse Reconstruction on Robust Dictionary for Accurate Salient Object Detection

Considering the low accuracy of existing salient object detection algorithms in addressing boundary objects problem, this paper proposes a novel algorithm which detects salient object through Sparse Reconstruction on Robust Dictionary. Firstly, the prior information of the salient objects connected with image boundary is statistically analyzed based on public datasets. Secondly, the evaluation and optimization of dictionary elements are performed to construct robust dictionary via removing low background boundary superpixels and adding high background internal superpixels. saliency maps are generated according to the sparse reconstruction errors on the new dictionary. Finally, integration of multi-level saliency maps is achieved by means of linear weighted fusion strategy with learned weight coefficients. Extensive experimental results demonstrate that the proposed algorithm is superior to the state-of-the-art models on benchmark datasets.