On seeing transparent surfaces in stereoscopic displays
Abstract (summary)
Transparency presents an extreme challenge to stereoscopic correspondence and surface interpolation, particularly in the case of multiple transparent surfaces in the same visual direction. The present work addressed the phenomenon of stereo-transparency in both psychophysical and computational domains. First, a series of psychophysical experiments were conducted, which investigated the limits of human ability to perceive stereo-transparency, as well as studied the interaction of density, disparity and the number of planes and its influence on observers' performance. Results of the study expand on earlier work in the area and account, for discrepancies in the existing literature. Next, three computational models of stereopsis that explicitly claim to be biologically plausible and to support transparency to some degree, were implemented. Their performance on transparent stimuli was compared with respect to each other and to the psychophysical data collected in the experimental part. Computational simulations provided an insight into the nature of the depth perception mechanism and a basis for quantitative evaluation of existing models.