Binocular stereo cues are critical for distinguishing 3D surface orientation, particularly at close range. For a study, researchers created a single-interval task in which observers were asked to discern the slant of a heavily textured planar test surface relative to a textured planar surrounding reference surface. Although the surfaces were shown in proper perspective, the stimuli were constructed in such a way that binocular signals dominated performance. Slant discrimination performance in the left and right eyes was assessed as a function of the reference slant and the amount of uncorrelated white noise introduced to the test-plane pictures. They contrasted human performance with a near-ideal observer (planar matching [PM]) and two sub-ideal observers.
For all conceivable slants, tilts, and distances, the PM observer utilizes the picture in one eye and back projection to anticipate a test image in the other eye. The forecast that most closely matched the measured image in the opposite eye determined the predicted slope, tilt, and distance. The first sub-ideal observer (local planar matching [LPM]) employed PM on local neighborhoods before pooling estimates across the test plane. Only location disparity is used by the second suboptimal observer (local frontoparallel matching [LFM]). The ideal observer (PM) and first sub-ideal observer (LPM) outperformed the second sub-ideal observer (LFM), illustrating the added benefit of pattern discrepancies.
Reference:jov.arvojournals.org/article.aspx?articleid=2778771
