Fine-scale processing in human binocular stereopsis.

Many studies have demonstrated that the human visual system is sensitive to very small differences in relative binocular disparity. It is not known over what monocular regions information is spatially integrated to mediate performance in such tasks. In this study we present psychophysical observations that define the smallest spatial scale involved in disparity processing, and we indicate the nature of the computations performed by the units mediating that disparity discrimination. We show that human observers can identify the sign of disparity of a single target dot when it is embedded in a row of identical dots, with these noise dots presented either in the fixation plane or with a proportion binocularly uncorrelated. In conjunction with the psychophysical data, we explore how a class of simple correlator models of stereopsis must be constrained in order to account for human performance for the same fine-scale tasks. Such models can perform the task only when the correlation is carried out over a very small region of the image, for a very small range of disparities. Our results demonstrate that there is a fine-scale input to the stereo system, mediated by foveal mechanisms that spatially integrate visual signals over a region as small as 4-6 arcmin in diameter.