Abstract
How local orientation information is pooled to yield global shape is a fundamental question in vision research. The aim of this study was to probe the characteristics of global shape mechanisms.
Stimulus arrays were oriented Gabors which sampled the circumferences of multiple concentric shapes in a polar grid. Individual Gabors were oriented tangentially to the global shape (signal) or randomly (noise); coherence was the independent variable. For signal elements positioned randomly in the array, about 15 signal elements (10%) were sufficient to detect concentric circular shapes. Sampling from ellipses, smoothed triangles, squares or pentagons yielded equally high sensitivities; performance declined for more complex shapes. Confining the signal to a single annulus reduced the number of elements needed to reach threshold to as few as 4 (2.5%). Overall performance for an array can be explained by probabilistic summation across multiple concentric shape detectors. We utilised the different rules found for signal summation within, compared to across, detectors to investigate the tuning of shape mechanisms. When elements were offset radially within 5% of the shape's radius, thresholds increased only slightly but doubled for 20% offsets. When element orientations were jittered within 12° of being tangent to the shape, thresholds remained constant but doubled for 25° jitter.
Our results provide evidence for highly sensitive detectors, tuned to circles but also other shapes, which sum information globally but only within specific annuli. These global mechanisms are broadly tuned with respect to the position and orientation of the local detectors from which they pool information. The relatively broad tuning for position makes an interesting prediction: if elements were taken from a pentagon and placed on a circular ring, they should be perceived as a pentagon rather than a jagged circle. Inspection shows this to be the case and provides a compelling new shape illusion.
This research is supported in part by EPSRC Grant No. GR/S59239/01.