Rules for combining the outputs of local motion detectors to define simple contours

We know something about the fidelity with which motion can be detected in local regions of the visual field but nothing about how these local motion signals are combined across space to define contours. To investigate such linking rules, we measured the detectability of motion-defined contours using an adaptation of the paradigm of Field, Hayes, and Hess (Vision Research, 33 (1993) 173) in which subjects are asked to detect the presence of simple contours defined solely by local motion direction that are embedded in a field of otherwise random local motions. We show that contours defined by motion whose direction is along the contour are more detectable than contours defined by motions of any common direction. Furthermore, the contour configuration is important in that straight and moderately curved contours, though not highly curved ones, can support this specialized form of motion integration.