Stereoscopic occlusion junctions

Abstract

Portions of surfaces in a binocularly viewed scene may be 'half occluded', that is, visible in only one eye. The human visual system uses zones of half occlusion to help segment the visual scene and infer figure-ground relationships at object boundaries. We developed a quantitative model of the depth-discontinuity cue provided by half occlusion. Half occlusions are revealed by two-dimensional interocular displacements of binocularly viewed occlusion junctions, such as T junctions. We derived a formula relating this two-dimensional displacement, or 'pseudodisparity', to binocular disparities and orientations of occluding and occluded contours. In human psychophysical experiments, perceived depth and contour orientation quantitatively depended on pseudodisparity, as predicted by our model, implying that the visual system senses quantitative variations in interocular junction position to reconstruct occlusion geometry.