The Importance of Natural Image Statistics in Visual Motion Estimation

Animals use visual signals to compute velocities in the environment.
In classic models, motion is computed only from pairwise stimulus correlations.
Here I argue that higher-order correlations are also important. First, I present
a theoretical formalism that relates correlational motion estimation to the statistics
of visual stimuli. The theory predicts that bright-dark asymmetries in natural environments
permit animals to improve motion estimation with odd-ordered correlations. To test
this idea, we adapted human psychophysics experiments to show that flies detect
motion in third-order correlations. Interestingly, the third-order correlations
that cause human and fly motion percepts are predictive of the motion of natural
scenes. Overall, our results suggest that natural visual stimuli influence motion
estimation in diverse species.