Abstract
In previous work, we found that as the inducing edges of a partly-occluded contour become non-relatable (i.e., can be interpolated only with an inflected contour), observers' interpolation settings of position and orientation become mutually inconsistent: there is no single smooth contour that they all agree with (Fulvio, Singh, & Maloney, CVPR2006). We propose that this inconsistency arises from the presence of multiple contours generated within the region of interpolation. We test this hypothesis by asking whether observers' settings at a given location are influenced by the presence of their own setting at a nearby location.
Methods: The experiment was run in two parts. In Part I, observers made paired settings of position and orientation independently through four interpolation windows. In Part II, two interpolation windows were opened on each trial: one contained a fixed line segment determined by the observer's own settings in Part I. The second contained the adjustable probe. The vertical offset between the two inducing edges was manipulated so that they were either relatable (R), just-relatable (JR), or non-relatable (NR).
Results: To test whether observers' interpolation settings within a window are altered by the presence of their own setting in a nearby window, we compared observers' settings in Part I versus Part II of the experiment. For relatable inducers, the settings were not reliably different between Part I and Part II (only 2/32 tests were significant for position). For the JR and NR inducers, however, a large majority of the positional settings were reliably different between the two parts (23/32 for JR, 24/32 for NR). This suggests that no single contour is interpolated in the JR and NR cases. Rather, multiple contours are present and, depending on where a measurement is taken, it is influenced more by one inducing edge or the other.
NSF BCS-0216944 NIH EY08266