Abstract
Classic computational models of stereoscopic organization incorporate spatial interactions to help in solving the correspondence problem of stereoscopic surface reconstruction in complex cluttered scenes. To evaluate the role of such spatial interactions in human vision, we used a task of detecting the correlation state of a pair of stereoscopically defined patches in a field of dynamic random dots that either became binocularly correlated (BC) to form a surrounding surface of the same duration as the target patches or remained binocularly uncorrelated (BU) to form a field of incoherent depth continuous with the pre- and post-test intervals. The target patches could be both BC, both BU or a mixed pair; each patch being BC or BU with 50% probability. The observer had to indicate whether the targets had the same or different correlation states, requiring attention to both targets across space in order to perform the task. Target pairs in sets of four orientations were presented in either the presence or absence of a simultaneous surround plane. The effect of the target pair orientation was not significant. Performance as a function of target duration was best when the surround plane was presented, allowing discrimination of the joint target state in as little as 30 ms. Importantly, removal of the surround plane degraded performance by a factor of 5–10. The results indicate a substantial role of the simultaneous surround plane in solving the stereo correspondence problem, even though it was presented for only the very brief duration of the targets and could not therefore reduce the uncertainty of the target correlation state except through the spatial interactions that are the subject of the study. Our findings extend the BU/BC transition anisotropy of Julesz & Tyler (1976) to the domain of spatial integration and 3D attention in stereoscopic space.