A visual mechanism in curve detection

To investigate how a smooth curve is processed in the human visual system, we introduce a new grouping principle explaining perceptual organization. Five observers were presented with arrays of band-pass elements, and the detection performance of an undulating path among randomly oriented distracters was measured as a function of the number of path elements and the orientation differences between adjacent path elements. We found that the undulating path was better detected when the number of path elements was large or the orientation differences were small and constant. Based on these results, we propose a model that groups the path elements to be perceived as a contour; local orientation differences are detected by curvature channels, the outputs of these channels are globally pooled with lateral inhibition, and the signals are detected by contour detectors of a threshould device.