An Image Sparse Representation for Saliency Detection

This paper presents a novel method for detecting saliency in static images based on image sparse representation. For each color channel, first, the image is partitioned into non-overlapping patches and each patch is represented by the way of sparse coding from a learned dictionary of patches from natural scenes. Then, global saliency and local saliency are calculated and fused to attain saliency of each patch. Local saliency is shown by popping out a patch from its surrounding patches. Global saliency is indicated by the rarity of a patch in the overall patches of the image. The final saliency map is attained by normalizing and fusing local and global saliency maps of all color channels. Experimental results in the benchmark image dataset demonstrate that the proposed method achieves a superior performance compared with most of state-of-the-art methods. Furthermore, both robustness and the low computational complexities make the presented algorithm feasible for subsequent applications.