Abstract
Human perception appears to incorporate the statistics of natural contours as they appear in images (Geisler et al., 2001). Neighboring line segments are constrained in direction, distance and orientation if they are part of the same contour. Last year we showed that observers primarily used segments near the axis of symmetry in discriminating a parabolic contour from a stimulus in which the line segments were individually rotated by 90 degrees forming a radially oriented pattern with the same distance and direction but differing orientation statistics. We have now conducted additional experiments a) controlling for factors that could have caused reduced orientation discriminability for more eccentric segments and b) using a circular contour with equal curvature at all points. The patterns were composed of 13 oriented segments, 0.17 deg in length with a center-to-center distance of 0.42 deg. On each trial uniformly distributed random orientation noise in a range of ± 90 deg was added to the stimulus, and the pattern was presented randomly within a range of distances around fixation at one of four possible orientations. Observers classified the pattern presented on each trial as a variant of the semicircular contour or its radial counterpart. After 5120 trials for each participant, logistic regression of the observers' responses on the orientation noise added to each segment of the pattern continued to show higher weightings near the axis of symmetry. Based on the data in Geisler et al. (2001), we computed the likelihood of neighboring segments on each trial belonging to the same physical contour. These likelihoods predict the responses for observers, although there are large individual differences. The higher weighting of segments near the axis of symmetry could reflect flanking facilitation from more eccentric segments on the grouping of center segments or possible attentional effects.