In this paper, we propose a system to obtain a depth ordered seg-\udmentation of a single image based on low level cues. The algorithm\udfirst constructs a hierarchical, region-based image representation of\udthe image using a Binary Partition Tree (BPT). During the building\udprocess, T-junction depth cues are detected, along with high convex\udboundaries. When the BPT is built, a suitable segmentation is found\udand a global depth ordering is found using a probabilistic framework.\udResults are compared with state of the art depth ordering and\udfigure/ground labeling systems. The advantage of the proposed ap-\udproach compared to systems based on a training procedure is the\udlack of assumptions about the scene content. Moreover, it is shown\udthat the system outperforms previously low-level cue based systems,\udwhile offering similar results to a priori trained figure/ground label-\uding algorithms
展开▼
