The demand for compute power and transmission bandwidth is growing rapidly as the display technologies progress towards higher spatial resolutions and frame rates, more bits per pixel (HDR), and multiple views required for 3D displays. Advancement in real-time rendering has also made shading incredibly complex. However, GPUs are still limited in processing capabilities and often have to work at a fraction of their available bandwidth due to hardware constraints.

In this dissertation, I build upon the observation that the human visual system has a limited capability to perceive images of high spatial and temporal frequency, and hence it is unnecessary to strive to meet these computational demands. I propose to model the spatio-temporal limitations of the visual system, specifically the perception of image artefacts under motion, and exploit them to improve the quality of rendering.

I present four main contributions: First, I demonstrate the potential of existing motion quality models in improving rendering quality under restricted bandwidths. This validation is done using an eye tracker through psychophysical experiments involving complex motion on a G-Sync display. Second, I note that the current models of motion quality ignore the effect of displayed content and cannot take advantage of recent shading technologies such as variable-rate shading which allows for more flexible control of local shading resolution. To this end, I develop a new content-dependent model of motion quality and calibrate it through psychophysical experiments on a wide range of content, display configurations, and velocities. Third, I propose a new rendering algorithm that utilises such models to calculate the optimal refresh rate and local shading resolution given the allowed bandwidth. Finally, I present a novel high dynamic range multi-focal stereo display that will serve as an experimental apparatus for next-generation of perceptual experiments by enabling us to study the interplay of these factors in achieving perceptual realism.