Reconstruction of Curved Surfaces from Random-Dot Stereograms

Various constraints have been used to recover 3D structures from 2D images. They are classi ed into two categories: explicit constraints and implicit ones. In stereo matching algorithms, smoothness constraint is an example of explicit constraints, because it explicitly speci es which binocular pairs are compatible and it forces, for example, to nd a planar surface from a pair of random-dot stereograms. As a result most algorithms using the smoothness constraint lack generalizing capability to nonplanar surfaces with non-zero curvatures. Implicit constraints have also been used in stereo matching algorithms. Ordering constraint is such an example. It does not specify which binocular pairs are compatible, but it only speci es which binocular pairs are not compatible. Therefore, it is an implicit constraint from the aspect of recovering surfaces.