Abstract
The three-dimensional structure of the world can be reconstructed using the differences, or binocular disparities, between the positions and appearance of the images of the same objects on the two retinae. Occlusion of one object by another gives rise to areas visible only in one eye, called monocular occlusions, for which binocular disparities cannot be computed. Nevertheless, monocular occlusions can be perceived at precise locations in depth and can even induce the perception of illusory occluding surfaces. Since a growing body of literature has shown that monocular occlusions are an integral part of stereoscopic depth perception, it is important that we understand the mechanisms of depth extraction from monocular occlusions. Psychophysical experiments suggest that the visual system is able to assign depth from monocularly occluded areas based on the constraints imposed by the viewing geometry. However, none of the existing models of stereopsis use viewing geometry as the primary mechanism for quantitative and qualitative depth extraction in occluded areas. Moreover, no model has been shown to recover depth and structure of illusory occluding surfaces induced by the presence of monocular regions. We propose a new model of depth perception from disparity and monocular occlusions in which monocularly occluded areas are detected explicitly and quantitative depth from occlusions is calculated based on occlusion geometry. The model represents several levels of processing in the visual cortex and includes complex interactions between disparity and monocular occlusion detectors. It successfully reconstructs depth in a large range of stimuli including random-dot stereograms, illusory occluder stimuli, da Vinci arrangements and natural images. Thus we demonstrate that a dedicated set of mechanisms for processing of monocular occlusions combined with classical disparity detectors can underpin a wide range of stereoscopic percepts.
Meeting abstract presented at VSS 2013