A Bayesian approach to half-occlusions M Zannoli, M Banks |
|---|
School of Optometry, University of California, Berkeley, CA, United States
|
| In natural scenes, distant surfaces are often occluded to one eye by nearby surfaces. Binocular disparity cannot be computed in these monocular regions, but those regions are nonetheless perceived at specific depths. To better understand how depth is estimated in this situation, we developed a probabilistic model of depth estimation with half-occlusions. The model incorporated probability distributions associated with occlusion geometry and a zero-disparity preference and distributions associated with the observed azimuth and blur of the monocular dot. We tested the model’s predictions in a set of experiments. In each experiment, a monocular dot was presented to the side of a binocular occluder. Participants indicated the perceived 3D location of the monocular dot by adjusting a binocular probe until the perceived azimuths and depths of the dot and probe were equal. We first asked whether the fixation distance relative to the occluding surface mattered to the perceived depth of the monocular dot. We found that it did; when the fixation distance was closer than the occluder, the perceived depth of the dot decreased and when the fixation distance was farther, the dot’s perceived depth increased. We next asked whether the sharpness of the monocular dot mattered. We found that it did not affect the average perceived depth but did affect the variance of depth settings with higher variance associated with greater blur. |
| Up Home |
|---|