Abstract
How does the visual system extrapolate contour shape? We investigated the influence of contour curvature on the shapes of visually-extrapolated contours and the precision with which they are represented. Observers viewed displays in which a contour segment disappeared behind an occluder, and they adjusted the position and orientation of a line probe protruding from the opposite side, in order to indicate the perceived location and tangent direction of the extrapolated contour. Measurements were taken at 6 different distances from the point of, thereby providing a relatively detailed representation of the perceived extrapolated shape. 9 different inducing contour segments were tested: 1 linear, 4 circular, and 4 parabolic (with 4 different values of curvature at the point of occlusion). The positional and orientational measurements for each contour were analyzed for variability, bias, and global consistency. The results demonstrated that the extrapolated shapes are localized fairly precisely (SDs in angular position < 10 degrees), with the precision nearly constant for each contour—in angular terms—with increasing distance from the point of occlusion. However, there was a clear cost of curvature, in that variability increased systematically with the curvature of the inducing contour. The results also demonstrated a strong influence of curvature on extrapolated shape, with extrapolated contours becoming systematically more curved with increasing inducer curvature. Although there were differences between observers in the precise shapes perceived, local measurements within each observer exhibited a high degree of internal consistency—with local settings of position and orientation at different distances from the point of occlusion being consistent with a global, smooth contour. We discuss implications for geometric and neural models of contour interpolation, and generative contour models that capture the visual system's distributional assumptions about contour shape.
Supported by NSF grant BCS-0216944.