Depth cue contributions in complex natural scenes

Perception of depth in natural viewing is based on information reported by many cues. These cues have been studied extensively individually, but many experiments in this area isolate a single cue, or a pair, in order to measure their relative reliabilities, and therefore weighting within the depth estimate under Bayesian cue combination. However, isolating cues does not simulate natural viewing, and studies show that simplifying viewing can make decoding a scene challenging for the visual system. Additionally, much work in this area presents overly-simplified stimuli, such as lines or dots, although these do not represent the complexities of natural objects, and are a poor fit for simulating the challenge posed to the visual system in decoding natural scenes. A branch of research has sought to address these issues of simplification of viewing and stimuli alike, by presenting more complex stimuli under naturalistic viewing conditions. However, the stimuli chosen, often photographs of smooth real or simulated shapes, often do not represent complex, naturally-occurring scenes, and report relative interactions rather than discrete weightings. The work presented in this thesis sought to address these issues by using advanced technological equipment and techniques to produce complex, naturalistic scenes with which to measure the contribution of depth cues to the weighted Bayesian estimate. The experiments contained within explore both retinal and extraretinal cues, binocular and pictorial cues, and a range of global and local viewing conditions. Overall, results show modest weightings for binocular cues such as disparity and vergence, little benefit to the overall weighting for the occluding contour and surface luminance pictorial cues, and a clear result of pictorial shape from shading in a real-world facial makeup study, with evidence throughout of benefits for the inclusion of complex naturalistic stimuli in future work of this kind.