Abstract
How do humans infer an object’s direction of motion from noisy sensory input? We hypothesized that observers utilize not only motion but also orientation information in their inferences, capitalizing on “streaks” created by moving objects (see also Geisler, 1999, Nature). We implemented this hypothesis in an ideal (Bayesian) observer framework, in which an observer’s knowledge is quantified using probability distributions. This led to several predictions that we tested using psychophysics and fMRI. Participants viewed dots moving coherently in a random direction and then reported the perceived direction of motion. Using a probabilistic pattern-based analysis (cf. van Bergen, Ma, Pratte & Jehee, 2015, Nature Neuroscience), we decoded the probability distribution of motion direction from activity patterns in visual areas V1-V4, and hMT+. Corroborating the predictions of the Bayesian observer model, we found that 1) probability distributions decoded from cortical activity had a characteristic bimodal shape, consistent with the notion that orientation might provide an important cue to the direction of motion. 2) The widths and 3) locations of the two peaks of the decoded probability distributions predicted, respectively, trial-by-trial variability in the participants’ behavioral responses, and the magnitude and direction of their behavioral errors. Finally, (4) in a follow-up behavioral experiment, analysis of the behavioral response distribution revealed a similar bimodal pattern, with one peak roughly centered at 0° and the other at 180° with respect to the true motion direction. Thus, observers sometimes perceived the stimulus as if it moved in a direction opposite to its true direction of motion, as predicted by the model. Together, these results suggest that human observers use not only motion, but also orientation information in their judgments of a moving stimulus, and moreover reveal the neural basis of the inference process involved.