1 Introduction
Reliably identifying the spatial extent of objects in images is important for high-level vision tasks like object recognition. A region that covers an object fully provides a characteristic spatial scale for feature extraction, isolates the object from potentially confusing background signals and allows for information to be propagated from parts of the object to the whole. For example, a region covering a human fully makes it possible to propagate the person identity from the easier to identify face area to the rest of the body.