Multicriteria saliency detection: a (exact) robust network design approach (Q2178371)

The paper proposes a framework for multiattribute robust saliency detection in image processing. The proposed method combines existing saliency detection methods and computes the salient region as the one with the minimal maximum weighted regret among the different saliency approaches. The authors propose a mixed integer programming formulation for this MRSD (multicriteria robust saliency detection) problem. They then survey several existing saliency detection approaches. They devise a new metod called JD that tries to identify image contour by measuring the local colour change. They compare the performance of MRSD and JD to existing methods in numerical experiments on five different datasets of more than 3000 images in total. They find that JD performs better than several existing methods and that MRSD provides robust solutions that perfrom well with respect to all of the saliency detection