Bayesian combination of local motion signals in human vision

Yair Weiss and Ted Adelson
UC Berkeley and MIT

In order to estimate the motion of an object, the visual system needs to combine multiple local measurements with varying degrees of ambiguity. We present a simple Bayesian estimator that calculates the posterior probability of a velocity given the local measurements and assuming a prior favoring slow speeds. In reviewing a wide range of previously published phenomena, we find that the predictions of the Bayesian estimator are in remarkable agreement with the percept of human subjects. This suggests that the seemingly complex set of illusions may be a result of a single computational mechanism that is optimal under reasonable assumptions.