Abstract

Purpose: It is commonly reported that the perceived direction of moving Type 2 plaids is biased away from the veridical intersection-of-constraints (IOC) direction and toward the vector-sum direction. A recent analysis of a Bayesian model of motion extraction predicts that when one of a plaid's component gratings gives a more reliable motion signal than the other component grating, perceived direction should be biased toward the grating with the more reliable motion signal. In Type 2 plaids, a sufficiently large bias toward the lower-speed grating could yield perceived direction beyond the vector-sum direction. By presenting Type 2 plaids whose components are of unequal spatial frequency (SF), we can manipulate the reliabilities of the gratingss' respective motion signals and thus test the prediction. Methods: Naïve subjects viewed Type 2 plaids whose component gratings drifted in directions separated by 15° and whose speeds differed by a factor of sqrt(1.5). We manipulated the ratio of the gratings' SFs while holding constant the geometric mean of their SFs. Subjects pointed an arrow in the direction of perceived drift. Results: When the gratings were of equal SF, perceived direction was close to the vector-sum direction. When the faster grating had higher SF, perceived direction was between IOC and vector-sum directions. When the slower grating had sufficiently higher SF, perceived direction was beyond the vector-sum direction, as predicted. Conclusions: A Bayesian computational model of motion extraction suggests that the vector-sum rule is a special case of an endpoint prediction whose general case is a weighted vector sum, with weights given by reliability of a grating's motion signal. Because a computational analysis indicates that the reliability varies with SF, the model predicts bias toward the higher-SF grating, which sometimes entails bias beyond the vector-sum direction. This prediction was confirmed.