Abstract
Binocular disparity is a very potent cue to depth, but is encoded with poor spatial and temporal resolution compared to luminance information. This fact is important for distributors of 3D content, since it opens up the possibility of saving bandwidth by transmitting a detailed luminance image + coarse depth map. I will discuss the cortical mechanisms believed to be responsible for the relatively poor resolution of stereo 3D. The initial step is the extraction of disparity in primary visual cortex. Here, neurons compute something close to the cross-correlation between image-patches in left and right eyes. Critically, they respond best to patches of uniform disparity. This is quite unlike luminance, where Hubel & Wiesel famously showed that neurons prefer step changes in luminance. This means that the scale with which we encode disparity is set by the size of V1 receptive fields themselves, not by the size of their subregions. Surprisingly, we encode complex contingencies between disparity and motion in transparent stimuli with the essentially same resolution as disparity in solid surfaces, suggesting that these must be important ecologically. Finally, at low frequencies we are more sensitive to horizontally-oriented variation in depth, as occurs in the ground plane, than to vertically-oriented variation, as occurs for example at a door jamb. I shall present new data and a speculation about why this is so.
Meeting abstract presented at OSA Fall Vision 2012