Human factors in the perception of stereoscopic images



Black, RH
(2017) Human factors in the perception of stereoscopic images. PhD thesis, University of Liverpool.

[img] Text
200193328_Jan2017.pdf - Unspecified

Download (24MB)

Abstract

Research into stereoscopic displays is largely divided into how stereo 3D content looks, a field concerned with distortion, and how such content feels to the viewer, that is, comfort. However, seldom are these measures presented simultaneously. Both comfortable displays with unacceptable 3D and uncomfortable displays with great 3D are undesirable. These two scenarios can render conclusions based on research into these measures both moot and impractical. Furthermore, there is a consensus that more disparity correlates directly with greater viewer discomfort. These experiments, and the dissertation thereof, challenge this notion and argue for a more nuanced argument related to acquisition factors such as interaxial distance (IA) and post processing in the form of horizontal image translation (HIT). Indeed, this research seeks to measure tolerance limits for viewing comfort and perceptual distortions across different camera separations. In the experiments, HIT and IA were altered together. Following Banks et al. (2009), our stimuli were simple stereoscopic hinges, and we measured the perceived angle as a function of camera separation. We compared the predictions based on a ray-tracing model with the perceived 3D shape obtained psychophysically. Participants were asked to judge the angles of 250 hinges at different camera separations (IA and HIT remained linked across a 20 to 100mm range, but the angles ranged between 50° and 130°). In turn, comfort data was obtained using a five-point Likert scale for each trial. Stimuli were presented in orthoscopic conditions with screen and observer field of view (FOV) matched at 45°. The 3D hinge and experimental parameters were run across three distinct series of experiments. The first series involved replicating a typical laboratory scenario where screen position was unchanged (Experiment I), the other presenting scenarios representative of real-world applications for a single viewer (Experiments II, III, and IV), and the last presenting real-world applications for multiple viewers (Experiment V). While the laboratory scenario revealed greatest viewer comfort occurred when a virtual hinge was placed on the screen plane, the single-viewer experiment revealed into-the-screen stereo stimuli was judged flatter while out-of-screen content was perceived more veridically. The multi-viewer scenario revealed a marked decline in comfort for off-axis viewing, but no commensurate effect on distortion; importantly, hinge angles were judged as being the same regardless of off-axis viewing for angles of up to 45. More specifically, the main results are as follows. 1) Increased viewing distance enhances viewer comfort for stereoscopic perception. 2) The amount of disparity present was not correlated with comfort. Comfort is not correlated with angular distortion. 3) Distortion is affected by hinge placement on-screen. There is only a significant effect on comfort when the Camera Separation is at 60mm. 4) A perceptual bias between into the depth orientation of the screen stimuli, in to the screen stimuli were judged as flatter than out of the screen stimuli. 5) Perceived distortion not being affected by oblique viewing. Oblique viewing does not affect perceived comfort. In conclusion, the laboratory experiment highlights the limitations of extrapolating a controlled empirical stimulus into a less controlled “real world” environment. The typical usage scenarios consistently reveal no correlation between the amount of screen disparity (parallax) in the stimulus and the comfort rating. The final usage scenario reveals a perceptual constancy in off-axis viewer conditions for angles of up to 45, which, as reported, is not reflected by a typical ray-tracing model. Stereoscopic presentation with non-orthoscopic HIT may give comfortable 3D. However, there is good reason to believe that this 3D is not being perceived veridically. Comfortable 3D is often incorrectly converged due to the differences between distances specified by disparity and monocular cues. This conflict between monocular and stereo cues in the presentation of S3D content leads to loss of veridicality i.e. a perception of flatness. Therefore, correct HIT is recommended as the starting point for creating realistic and comfortable 3D, and this factor is shown by data to be far more important than limiting screen disparity (i.e. parallax). Based on these findings, this study proposes a predictive model of stereoscopic space for 3D content generators who require flexibility in acquisition parameters. This is important as there is no data for viewing conditions where the acquisition parameters are changed.

Item Type: Thesis (PhD)
Divisions: Faculty of Health and Life Sciences > Institute of Population Health
Depositing User: Symplectic Admin
Date Deposited: 14 Aug 2018 08:48
Last Modified: 01 Nov 2021 08:14
DOI: 10.17638/03014678
Supervisors:
URI: https://livrepository.liverpool.ac.uk/id/eprint/3014678