Saliency detection based on background seeds by object proposals and extended random walk

Recently, many graph-based algorithms are applied in the research of saliency detection, which use the border of an image as a background query. This frequently leads to undesired errors and retrieval outputs when the boundaries of the salient objects concerned touch, or connect with, the image's border. In this paper, a novel bottom-up saliency-detection algorithm is proposed to tackle and overcome the above issue. First, we utilize object proposals to collect the background seeds reliably. Then, the Extended Random Walk (ERW) algorithm is adopted to propagate the prior background labels to the rest of the pixels in an image. Finally, we refine the saliency map by taking both the textural and structure-information into consideration simultaneously. Experiments on publicly available data sets show that our proposed method achieves competitive results against the state-of-the-art approaches.