Saliency Cuts on RGB-D Images

Saliency cuts aims to segment salient objects from a given saliency map. The existing saliency cuts methods focus on dealing with RGB images and videos, but ignore the exploration of depth cue, which limit their performance on RGB-D images. In this paper, we propose a novel saliency cuts method on RGB-D images, which utilizes both color and depth cues to segment salient objects. Given a saliency map, we first generate segmentation seeds with adaptive triple thresholding. Next, we extend GrabCut by combining depth cue, and use it to generate a roughly labeled map. Finally, we refine the boundary of the salient object adaptively, and produce an accurate binary mask. To the best of our knowledge, this method is the first specific saliency cuts method for RGB-D images. We validated the proposed method on the largest RGB-D image dataset for salient object detection, named NJU2000. The experimental results demonstrate that our method outperforms the state-of-the-art methods.