Abstract
Visual representations are designed to efficiently encode visually presented information that is used to guide actions and inform decisions. Visual representations of shape must efficiently encode shape geometry, but they must also support important behaviors, like comparisons across viewpoint, recognition under partial occlusion, and judgments of similarity. Here we propose a representation of 2D contour shape based on joined segments of constant curvature. This representation can efficiently encode contour geometry while also supporting important visually guided behaviors. In three experiments we demonstrate that shapes formed from constant curvature segments are better recognized under viewing conditions that require efficient storage than similar shapes that are not formed from constant curvature segments. Together, these experiments support the idea that encoding contour shapes as joined segments of constant curvature is a strategy used by the visual system for efficiently encoding these shapes in visual working memory. We also demonstrate, however, that under conditions that do not require efficient storage, shapes formed from constant curvature segments are no easier to recognize than similar shapes that are not formed from constant curvature segments. Specifically, performance differences only arise when shapes must be stored for more than 500 ms and compared from different viewpoints. Finally, we model how a constant curvature, parts-based representation trades off between fidelity and efficiency in the encoding of contour shape.