Stereo matching, i.e., the matching by the visual system of corresponding parts of the images seen by the two eyes, is inherently a 2D problem. To gain insights into how this operation is carried out by the visual system, we measured, in human subjects, the reflexive vergence eye movements elicited by the sudden presentation of stereo plaids. We found compelling evidence that the 2D pattern disparity is computed by combining disparities first extracted within orientation selective channels. This neural computation takes 10–15 ms, and is carried out even when subjects perceive not a single plaid but rather two gratings in different depth planes (transparency). However, we found that 1D disparities are not always effectively combined: When spatial frequency and contrast of the gratings are sufficiently different pattern disparity is not computed, a result that cannot be simply attributed to the transparency of such stimuli. Based on our results, we propose that a narrow-band implementation of the IOC (Intersection of Constraints) rule (; ), preceded by cross-orientation suppression, underlies the extraction of pattern disparity.
展开▼
