Purchase this article with an account.
Alan Lap-fai Lee, Alan Yuille, Hongjing Lu; Superior perception of circular/radial than translational motion cannot be explained by generic priors. Journal of Vision 2008;8(6):31. doi: https://doi.org/10.1167/8.6.31.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
A fundamental question in motion perception is how the visual system integrates local motion information over space to form a coherent global percept. We examined whether different motion flows affect spatial integration in motion perception, and whether a Bayesian model that includes generic but not specific priors is able to account for human performance. We adopted the stimulus developed by Nishida et al. (2006) to compare human performance for three types of motion flow: translational, circular and radial motion. Each stimulus consisted of 728 Gabor elements (drifting sine-wave gratings windowed by stationary Gaussians) in a 12° circular window. The orientations of the Gabors were randomly assigned and their drift velocities were determined by a specified global motion flow pattern. The motions of signal Gabor elements were consistent with global motion, but the motions of noise Gabor elements were randomized. Motion sensitivity was measured by the coherence threshold: the proportion of signal elements that yielded a performance level of 75% correct in a discrimination task of determining the global motion direction. We found worse performance for perceiving translation than circular/radial motion (coherence threshold: translation 44%, circular 34%, radial 28%). Such superior performance in circular/radial motion was also replicated in a detection task. We implemented a Bayesian model incorporating a generic prior preferring slow and spatially smooth motion. This model predicts slightly worse discrimination performance for circular/radial motion than for translation (coherence threshold: translation 35%, circular 38%, radial 39%), which is the opposite from human performance. Similarly, the model predicts the opposite trend from humans in the detection task. The failure of the model with generic priors suggests a specific prior for circular/radial motion is involved at the computational level. The apparent need to postulate special mechanisms is consistent with physiological evidence that MST neurons are selective to circular/radial motion.
This PDF is available to Subscribers Only