Abstract
When an object's component parts are not readily perceived as a linked whole due to occlusion, the visual system must perceptually group the disparate elements into a single coherent unit. Several principles of perceptual grouping, with varying influences on object formation, have been proposed and yet our understanding of the differences in the representation arising from different grouping cues remains rather poor. We adapted the psychophysical reverse correlation technique to measure the percept of two partially occluded objects formed by the grouping of common elements. Stimuli contained two perceptual rectangles to which we added orthogonal occluders, creating four discrete surface patches that could be grouped in two different ways. Manipulation of intrinsic parameters of each linking cue biased grouping in one direction. For example, during color similarity linking, two patches were light gray while the remaining two were dark gray, promoting the percept of two distinct gray rectangles. A second condition (blocked by session) was included for each linking cue in which the occluders were positioned “below” the rectangles (i.e., the rectangles were not occluded). These displays comprised the base images to which Gaussian distributed noise was added on each trial. Subjects determined the orientation of the perceived rectangles and classification images for each condition were created by aggregating noise images across trials. Results yielded access to an approximation of subjects' internal representation of the base images. Similar sectors of the display contained statistically significant pixels in both occluded and non-occluded conditions, suggesting that the underlying representations of the percepts were considerably congruous. Critically, in the occluded condition each subject showed diagnostic pixels located on occluded portions of the rectangles. However, since no information about the rectangles was actually present in these occluded regions, this provides evidence that perceptual grouping is used to demark surface patches into perceived objects.
Funded by NIH RO1-MH54246.