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 hal
f occlusion to help segment the visual scene and infer figure-ground relati
onships at object boundaries. We developed a quantitative model of the dept
h-discontinuity cue provided by half occlusion. Half occlusions are reveale
d by two-dimensional interocular displacements of binocularly viewed occlus
ion junctions, such as T junctions. We derived a formula relating this two-
dimensional displacement, or 'pseudodisparity', to binocular disparities an
d orientations of occluding and occluded contours. In human psychophysical
experiments, perceived depth and contour orientation quantitatively depende
d on pseudodisparity, as predicted by our model, implying that the visual s
ystem senses quantitative variations in interocular junction position to re
construct occlusion geometry.