Encoding and perception of 3D orientation




Shields, Stephanie M.

Journal Title

Journal ISSN

Volume Title



A fundamental task of the visual system is to construct a three-dimensional (3D) representation of the environment from the left and right eyes’ two- dimensional (2D) retinal images. The differences between those images are governed by projective geometry — the mapping of the 3D environment to the two retinae — and are used by the visual system to more precisely estimate depth. As such, one might expect that projective geometry has shaped how the visual system extracts and represents 3D information. Recent work suggests that is indeed the case for 3D motion perception and, moreover, demonstrates that combining existing knowledge of monocular responses to 2D stimuli with a geometry-focused modeling framework can explain noncanonical patterns in binocular responses to 3D stimuli (Bonnen et al., 2020). To both test whether such a framework generalizes to other domains and to advance knowledge of the processing of another 3D scene feature, this dissertation applies a projective-geometry-based framework to the study of 3D orientation, a likely important substrate of 3D shape computations. First, I describe the geometry of 3D orientation, parameterized as slant and tilt, and I establish a projection model that estimates the retinal orientations produced by a slanted line in the environment. I show that the disparity between those orientations could serve as a cue to 3D orientation and demonstrate the impacts of various changes in stimulus parameters. Second, I construct a theoretical neural population model based on existing knowledge of V1 responses to 2D orientation. I use this model to show how slant may impact V1 responses and to generate predictions about patterns that may be present in psychophysical performance. Third, I use V1 response data from Bridge and Cumming (2001), which shows that some binocular neurons have unequal monocular orientation preferences, to build a more directly data-informed population model, and I assess how the recorded neurons’ responses may vary with slant. Finally, I test the ability of human observers to discriminate between slants under various conditions. Perceptual performance generally follows the patterns predicted by the theoretical neural model and varies notably with the cues available to observers. Collectively, this work adds to existing understanding of the encoding and perception of 3D orientation and supports the importance of taking the environment-to-retina transformation into account.



LCSH Subject Headings