The statistics of natural binocular images
In recent years, the study of “natural” images has generated much interest in vision science. Of all the images that it would be possible to produce, those that one might expect to encounter outside of a psychology laboratory represent a tiny fraction, sharing important statistical properties. Studying these properties is likely to prove useful in understanding how the brain represents visual information. In this talk, I will discuss the known statistical properties of natural binocular images, and show how our best computational model of binocular processing in the primary visual cortex (the binocular energy model) responds to a set of such images. I will then review the evidence that (i) the encoding of binocular information at a single cell level and (ii) some of the known peculiarities of depth perception may be understood as an optimal encoding and interpretation of this information.