Purchase this article with an account.
Aaron Clarke, Stéphane Rainville; Motion grouping/segmentation via velocity gradients. Journal of Vision 2008;8(6):24. doi: 10.1167/8.6.24.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Introduction: Plants and animals bear extended limbs whose velocities vary smoothly with distance from the trunk or torso. Conversely, local velocity measurements change abruptly between occluding and occluded objects. This suggests velocity gradients may be useful for perceptual grouping/segmentation. Here we test perceptual grouping/segmentation as a function of velocity gradient magnitude.
Methods: Stimuli were obliquely oriented velocity gradients (either +45° or −45°) over dense random noise patterns with velocities either collinear or orthogonal to the gradient. Subjects indicated gradient orientation in a 2AFC task over multiple gradient magnitudes. Velocity endpoints (2 and 3 cycles/sec) and spatial frequency (2 cpd) were held fixed while gradient center location and gradient sign varied over trials.
Results: Performance for gradient-orientation identification peaks for steep velocity gradients and falls off smoothly for shallower gradients. Furthermore, performance is best when motion and gradient directions are orthogonal and worst when they are collinear.
Conclusions: A gradient's maximum and minimum velocity endpoints provide unambiguous orientation identification information, however, observers' performance varies significantly depending on whether the gradient between velocity endpoints is steep or shallow. This suggests that the visual system links similar local velocity estimates and segregates differing estimates, which subsequently makes the gradient endpoints difficult to distinguish in the shallow gradient case and easier to distinguish in the steep gradient case. While steep gradients generally facilitate perceptual segregation, motion directions orthogonal to the velocity gradient direction maximize this percept.
This PDF is available to Subscribers Only