Saliency model based on a discrete centre-surround

A novel saliency model based on a discrete centre-surround (C-S) is proposed. By addressing the problem of filter scales selection commonly existing in the state-of-the-art saliency methods, the contrast map, obtained by a discrete C-S, can uniformly highlight salient objects. Furthermore, a saliency probability measure is derived from the combination of the discrete C-S and saliency priors, which further enhances the object's saliency. Experimental results on a dataset containing 1000 test images with ground truths demonstrate that the proposed model clearly outperforms previous saliency models.