Abstract
Previously, we presented work suggesting that the contour integration evident in path detection displays (Field, Hayes, Hess, 1993) is really revealing the intermediate output of the first stage of the contour interpolation process: a stage in which contours are interpolated indiscriminately across all pairs of relatable edge fragments. In a second stage, interpolated contours are evaluated in the face of information from a variety of scene cues – if the information from scene cues supports the existence of the interpolated connection, the connection will be maintained or strengthened. If it runs contrary to the interpolated connection, the connection will be weakened or deleted. Here, we utilize a novel paradigm, inspired by a figure from Bregman (1990), to measure the effects of cues in the second stage and to determine how information from multiple cues is integrated. Our 2AFC task utilizes stimuli consisting of fragments of randomly arranged alphanumeric characters; hidden within this noisy background is a subset of fragments that form a whole number or letter if the gaps between the fragments are filled in with interpolated connections. The task requires participants to find and recognize the whole letter or number. Doing so requires building up a complete object description from many real and interpolated edges. The results suggest that cues as to border ownership, and amodal surface spreading (both luminance contrast polarity and equiluminant color contrast) play a role in determining the final perceptual strengths of interpolated contours in stage 2. In addition, the results support the possibility that interpolation proceeds separately in different spatial frequency channels and that corroboration across channels is taken into consideration to determine the perceptual weights of interpolated contours in stage 2. Finally, the results reveal that, somewhat surprisingly, information from multiple cues is integrated in a simple additive fashion, with one interesting exception.