Abstract
Part-based approaches organize global shape in terms of segmented parts and their spatial relationships. By separating the representation of parts from that of their spatial relationships, they provide a shape representation that is robust under transformations such as articulating limbs that are common in biological objects. Skeleton or axis-based approaches can provide a compact representation of both parts and their spatial relationships (Feldman & Singh, 2006; Singh & Feldman, VSS07). The current study measured visual sensitivity to different kinds of shape transformations based on axes and parts. In Expt 1, stimuli were simple elongated shapes, and four types of axis-based transformations were applied: length change, width change, curvature change, and orientation change. In Expt 2, the simple shapes used in Expt 1 were added to a base shape; hence each now constituted a part on a larger shape. The same four shape transformations were applied, plus a lateral shift in the location where the part protruded from the base. Observers saw a test shape (masked) followed by two successive alternatives (also masked). One of the two alternatives matched the test shape, the other was modified along one of the transformation dimensions, allowing us to measure thresholds for detecting the various types of shape transformations. In order to compare thresholds across transformations, they were converted into a common measure based on the normalized area of the symmetric difference. In both experiments, the highest sensitivity was found for changes in width and length of the part, followed by curvature, and then orientation. The sensitivity to lateral shifts of the part (in Expt 2) was the poorest. The results indicate that observers are most sensitive to changes involving the intrinsic parameters of a single axial branch, and less so to changes involving the spatial relationship between two axial branches.
NSF DGE 0549115 (Rutgers IGERT in Perceptual Science) NSF CCF-0541185.