Testing the Bayesian model of perceived speed.

In a recent Bayesian model by Weiss, Simoncelli, and Adelson, motion perception is biased by a prior favoring slow speeds. This model predicts qualitatively an impressive variety of phenomena, including the dependence of perceived speed on contrast. We show that the model can also generate quantitative predictions: for a drifting grating with contrast c, perceived speed is proportional to c(q)/(k(q) + c(q)), with k, q constants. We tested this expression on measurements of perceived speed as a function of contrast. Observers indicated the slower of two drifting gratings, a test and a standard. For each test contrast we found the test speed that appeared to match the standard speed. The model fits the data, but only if q is less than 2, the value it would have if the internal representation of contrast were linear. The Bayesian model can make correct quantitative predictions, but needs to be extended to incorporate a more realistic, nonlinear representation of contrast. Copyright 2002 Elsevier Science Ltd.