On Binocularly Viewed Occlusion Junctions

Under binocular viewing of a scene, there are inevitably regions seen only in one eye or camera. Normally, this is a source of trouble for stereopsis algorithms which must deal with these regions on non-correspondence. This paper points out that in fact half-occlusion can be a source of valuable information. This is done by deriving a formula relating the displacement of an occlusion junction in the two eyes' images and the depth difference between the scene edges that comprise the occlusion junction. This paper represents the first quantitative result on the cue of half-occlusion in stereopsis.