One of the most remarkable characteristics of the human visual system is its ability to perceive specific three-dimensional forms in single two-dimensional contour images. Several investigators have attributed this to an innate preferential bias in the visual system toward 'simpler' structures. I shall first present some work that seeks to formalize the intuitive notions of simplicity in a constrained domain of polyhedra. The resulting computational system is able to recover perceptually-correct 3D structures from several line drawings belonging to the domain. It does, however, encounter problems in trying to transcend domain boundaries.
An alternative, and perhaps more general, approach proposed by the early empiricists suggests that this ability may also be acquired from visual experience, with the three-dimensional percept being the manifestation of a learned association between specific two-dimensional projections and the correlated three-dimensional structures. I shall present an experimental paradigm that enables objective verification of this hypothesis by rendering the learning to be perceptually manifest. The results implicate high-level recognition processes in the task of shape-perception.
We thus have evidence for the visual system employing both bottom-up and top-down strategies for accomplishing the task of 3D shape-perception. Our current efforts focus on computationally modeling top-down influences on other supposedly 'early' perceptual tasks. We seek, in effect, to blur the distinction between visual cognition and perception.