Purchase this article with an account.
Reuben Rideaux, Andrew E Welchman; But still it moves: static image statistics underlie how we see motion. Journal of Vision 2020;20(11):275. doi: https://doi.org/10.1167/jov.20.11.275.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
How does experience shape what we see? Bayesian theories of vision assume that we systematically accumulate information about the statistical probability of particular events in the environment. A classic example being the use of a ‘slow world’ prior that is premised on close-to-zero net motion of the environment. Specifically, humans are thought to have internalised knowledge of environmental motion statistics, and to use this prior information to shape and guide perceptual estimates, thereby accounting for perceptions, and misperceptions, of movement. Here we explore these ideas by systematically manipulating the training inputs to a neural network to discover and test the drivers of motion perception.
We find that diverse motion characteristics are largely explained by the statistical structure of natural images, rather than motion per se. In particular, we show how neural and perceptual biases for cardinal motion directions result from the orientation structure of natural images. We reveal a novel interdependency between speed and direction preferences in MT neurons and show how this is explained by the autocorrelation in natural images. We demonstrate that motion illusions spontaneously emerge from the necessity to estimate the velocity of natural image sequences, and show how these can all be computationally explained within a biologically plausible system. Finally, we demonstrate that speed and image contrast are related quantities, and using behavioural tests, we show that it is knowledge of this speed-contrast association that explains why observers underestimate the speed of low contrast image sequences; that is, rather than the distribution of movements in the environment (i.e., the ‘slow world’ prior) as premised by Bayesian accounts. Together we provide an exposition of motion speed and direction estimation by biological brains, and offer concrete predictions for future neurophysiological experiments.
This PDF is available to Subscribers Only