Abstract: Diffusion-based saliency detection is one of many methods about image saliency detection.
The key is how to choose these seeds and propagate their saliency. Given the unconspicuous
difference of color and contrast between the background and the salient object, the erroneous seeds
may be selected. Finally, these wrong seeds could affect the quality of saliency map. In this paper, we
propose a novel seeds detection approach that take advantage of saliency map. By taking regularized
random walk and manifold ranking into account, the saliency map of an image is then estimated.
Experimental result demonstrates the proposed method performs well and provide pixel-wised
saliency map.

Key words: saliency, seeds, diffusion, propagation