Abstract
The visual system represents partially occluded surfaces by filling in the missing/occluded surface segments in the back and integrating them with the visible surface segments. However, because the missing segments are not imaged on the retina, the underlying 3-D amodal surface integration process must rely on the visual system's internal knowledge of the real world. The challenge is to discover the nature of the knowledge. Here, we first revealed the amodal integration process uses knowledge that segments of a unitary surface have the same contrast polarity. Our experiment used a stereo display that rendered a horizontal rectangle been perceived as occluded by two vertical bars. Due to occlusion, the retinal images of the horizontal rectangle were broken into a middle segment sandwiched by the vertical bars and two outer segments. We found that segments with the same contrast polarity amodally integrated leading to the perceived depth of the middle segment with zero binocular disparity adopting the depth of the two outer segments with uncrossed disparity. But with opposite contrast polarity, the segments were perceived at different depths indicating a failure of amodal integration. Next, we manipulated the 3-D boundary contour alignment and 3-D surface curvature of the visible segments, and revealed that the amodal integration process also uses knowledge of natural surface being smooth. Finally, we investigated the influence of the occluding images (vertical bars). We found amodal integration became weaker with increasing widths of the occluding bars. Furthermore, whereas T-junctions formed between the vertical bars and the horizontal segments are critical for 2-D amodal integration, their significance reduces in favor of having a common binocular depth assignment for the visible segments in the 3-D display. Overall, our study reveals the visual system internalizes knowledge of real world surfaces and geometric relationships to represent partially occluded surfaces in 3-D space.
Meeting abstract presented at VSS 2016