Abstract
Future 3-D displays are expected to effectively support full parallax visualization which would allow the viewer to experience a stereoscopic 3-D scene from multiple continuous perspectives. However, the optimal parallax density with respect to the human visual system is not yet known. In this study, we address this issue by analyzing the artifacts that arise in an image sequence replicating the changing views of a 3-D scene for a moving observer. We aimed to specify the minimal step in view density that is noticeable by measuring the visibility of the related artifacts. We simulated these effects on a passive stereo 3-D display and performed psychophysical studies to define a perceptual threshold for continuous parallax. In a 2IFC task, subjects indicated the "better looking" interval between a reference image sequence where the correct view was explicitly updated at every frame, and a distorted one where an angular sub-sampling factor was modulated to increase/decrease the severity of the artifacts. A light leakage among adjacent views was modeled in the spatial domain using a super-Lorentzian curve to account for additional display-related effects. We examined various synthetic stimuli and evaluated state-of-the-art, full-reference quality metrics in an attempt to model this behavior. We found that the top-performing metrics can be used to indirectly characterize the artifacts' visibility despite not being specifically constructed to capture distortions in the angular domain.
Meeting abstract presented at VSS 2018