Abstract
Subjects perform in the hyperacuity range when discriminating circles from Radial Frequency (RF) patterns: contours containing radial sinusoidal modulations. This high sensitivity results from integrating information across the entire contour. Points of maximum curvature have been implicated as a key feature in this integration process. To determine the role of local curvature in global shape processing, we measured curvature discrimination for complete contours (circles, RF3, 5 & 8; size=0.5deg; curvature=2.0/deg), isolated RF cycles and circular arcs (angular extents from 22.5-360deg). The task was to detect deviations from circularity (RF & isolated cycles) or the curve with higher curvature. For circular arcs, thresholds increase with increasing angular extent following a power-law relationship (slope = -0.56) but only up to 180deg (semi-circle). Extending the size further does not improve performance. Thresholds for discriminating circular segments (of half the angular extent of an RF cycle) were comparable to those for complete RF contours. Sensitivity for single RF cycles was significantly poorer than for complete RFs or circular arcs. Thresholds for RF arcs centred at points of convex contour maxima, contour minima and points of inflection showed no significant difference. Given that curvature thresholds for entire RF contours do not significantly exceed those for circular segments, one might speculate that RF computation is limited by detecting points of maximum curvature. This is, however, inconsistent with substantially reduced sensitivity for isolated RF cycles. Instead, results suggest that the signal of local curvature units elicited by RF patterns is degraded compared to circular segments and that global pooling of such sub-ideal local signals may explain the hyperacuity performance for these shapes. This reconciles an apparent paradox: RF sensitivity can be similar to that of isolated circular arcs and yet high sensitivity for RF contours results from global pooling of local information.
Meeting abstract presented at VSS 2013