UC San Diego

Nominally, the stereoscopic representation of a scene requires twice as much bandwidth as the corresponding monocular representation, since a separate view is presented to each eye. This increase in data rate is one of the most imposing challenges for stereoscopic media transmission and consumption. Mixed resolution stereoscopic coding (MRSC) has been proposed to mitigate the bandwidth requirements by transmitting a stereo pair comprised of one full resolution image and one lower resolution image. MRSC preserves the overall video quality by relying on the theory of binocular suppression, a perceptual phenomenon where if one eye's view of the world is blurry while the other eye's view is sharp, then the fused 3D percept of the scene will appear relatively sharp and faithfully represented in depth. This dissertation examines the binocular perception of blur and its application to 3D video processing. We begin by investigating the temporal aspects of MRSC---the effects of frame rate and blur on perceived quality and viewer fatigue---and whether balancing blur between both eyes is necessary. Subsequently, we develop a super-resolution method for MRSC that restores high frequency content to the low resolution half of the stereo pair. Our method results in sharper images and temporally consistent video. As part of our analysis, we derive expressions that quantify how much motion can aid the stereo matching process as a function of image features, noise, and motion distribution. These limits are relevant to the design of algorithms for spatio-temporal disparity estimation. Finally, in consideration of the visual systems's natural preference for horizontal disparity, we examine the influence of anisotropic spatial filtering on the perception depth. We obtain contrast sensitivity functions for depth detection as a function of filter orientation, and develop a computational model of sensitivity based on the power spectrum of the stimulus. These results have implications for theories of stereopsis mechanisms, and can also be used in rate distortion decisions for MRSC. Taken together, this work establishes MRSC as a perceptually plausible coding technique for stereoscopic video, and more generally, can be applied to the development of stereo algorithms and metrics