Abstract

Recent work in salient object detection has consideredthe incorporation of depth cues from RGB-D images. In most cases, depth contrast is used as the main feature. How-ever, areas of high contrast in background regions causefalse positives for such methods, as the background fre-quently contains regions that are highly variable in depth.Here, we propose a novel RGB-D saliency feature. Local Background Enclosure (LBE) captures the spread of angu-lar directions which are background with respect to the can-didate region and the object that it is part of. We show thatour feature improves over state-of-the-art RGB-D saliencyapproaches as well as RGB methods on the RGBD1000 and NJUDS2000 datasets.