Random dot stereograms and the correspondence problem revisited

Summary form only given, as follows. The random-dot stereogram is an important technique in vision research for portraying stimuli that are visible to neither eye alone but are visible in the binocular view. This lack of information in the real-world sense suggests that stereovision in humans might really be a low-level vision task not requiring extensive prior learning or intervention of higher visual-cortical areas. Different configurations of random-dot stereograms are used to determine as precisely as possible the mode of functioning and the usable limits of the stereo technique employed by the human visual system. The most significant problem that any such system faces is the stereo-correspondence problem. The difficulty is further compounded if the images presented to the vision system do not have any information which can be analyzed by the higher processing centers in the brain, but are composed of a matrix of randomly attributed image points. Based on the insights gained from experiments with random-dot stereograms, a neurally inspired stereovision system design is proposed whose hardware implementation, with components as slow as the neural counterparts, is capable of creating depth maps of input images in real time. This study also describes some of the results obtained from the software simulation of the proposed system design.<>