Abstract
Shape completion is essential for representing the visual world under conditions of partial occlusion. A number of accounts of shape completion have been proposed, some of which entail differences between modal (completing in front) and amodal (completing behind) cases. Our aim in this study was to better understand shape completion in general, to clarify what differences, if any, exist between modal and amodal completion, and specifically to understand how probabilistic estimation of the visual interpolated contour might differ in the two cases. Stimuli consisted of a red circle to which Gaussian noise was added at three levels of standard deviation. Overlaid on the circle was a dark triangle, and the circle was set up so that across all conditions the inducers' location and orientation was fixed. The circle was shown in either positive disparity (modal condition), or negative disparity (amodal condition) relative to the triangle. The visually completed shape perceived by the subject was measured by flashing a dot on top of the inducer, after which the subject had to indicate whether the dot appeared inside or outside the circular shape. A staircase procedure was used to find the distribution over possible perceived completions for each condition. The stimulus setup allowed us to readily compare different completion models. We found that the distribution over completions differed across different levels of standard deviation, even though the geometry of the local inducer was held constant, meaning that shape completion was influenced by statistical aspects of the global shape. For most subjects, we replicated the strong tendency for the perceived completing contour to be pulled "inward" towards the center of the circle. Overall, the results cannot be explained by conventional completion models, and suggest that modal and amodal completion processes embody somewhat different probabilistic assumptions about shapes.
Meeting abstract presented at VSS 2015