Several models have been proposed for how the brain measures velocity from
the output of motion-energy units. These models make some unrealistic assum
ptions such as the use of Gabor-shaped temporal filters, which are non caus
al, or flat spatial spectra, which are invalidated by existing data. We pre
sent a Bayesian model of velocity perception, which makes more realistic as
sumptions and allows the estimation of local retinal velocity regardless of
the specific mathematical form of the spatial and temporal filters used. T
he model is consistent with several aspects of speed perception, such as th
e dependence of perceived speed on contrast. (C) 2000 Elsevier Science Ltd.
All rights reserved.