Salient object detection based on sparse representation with image-specific prior

This paper presents a bottom-up salient object detection algorithm based on sparse representation with image-specific prior. First, we obtain image-specific prior by generating convex hull of interest points to estimate likely position of the salient object. Second, we construct background dictionary and object dictionary based on image-specific prior. For each image region, we compute reconstruction errors using sparse representation with respective dictionaries. Third, the pixel-level saliency score is measured by comparing two reconstruction errors. Experimental results on the MSRA-1000 dataset show that the proposed algorithm is competitive with recent state-of-the-art algorithms in terms of efficiency and accuracy.