Abstract
Normal viewing of a surface that lies behind another, and is partially occluded, results in each eye seeing parts that are invisible to the other. This seldom leads to binocular rivalry; instead one sees a stable surface behind the occluder. We asked whether this results from the visual system's ability to piece together a continuous binocular view from distinct segments seen by each eye alone. We constructed a textured surface from filtered 2D white noise. Observers viewed this via a stereoscope, with a binocular disparity that placed the surface behind a circular aperture. Vertical bars, with separation and width equal to the disparity, occluded the surface, causing alternate strips of the texture to be seen by one or other eye alone. Despite this, the texture appeared as a continuous surface occluded by vertical bars. No coherent surface was seen when the texture strips were binocularly discontinuous. To understand the conditions under which binocularly continuous surfaces appeared coherent, we measured how well observers could distinguish continuous from discontinuous surfaces, for textures varying in orientation and spatial frequency in the presence of different occluding surfaces. We call this measure surface coherence. This was highest for horizontal textures at lower spatial frequencies, and disappeared for all vertical textures. The weakest coherence is associated with the weakest binocular correlation in surface properties. When the disparity of the textures was inconsistent with real world occlusion, surfaces lacked coherence, even those with high binocular correlation. Our results show that stable perceptions of coherent surfaces can be pieced together from the two eyes' separate views, given two conditions. First, the image in one eye must predict what the other eye sees. Second, the arrangement of the monocular regions must be consistent with real world occlusion.
This work was supported by EY04440 and EY13079.