Abstract
Half-occluded zones (visible from only one eye) are found at every depth discontinuity in daily visual scenes. Even though such zones have no counterpart in the other eye (thus no disparity defined), they are perceived at a certain depth behind the occluding surface rather than causing binocular rivalry. Here we propose a mechanism detecting interocularly unpaired zones in each eye modeled after physiological responses of disparity selective cells and show a stereo algorithm that reconstructs 3D structures from not only interocularly paired but also unpaired points. In our model, we assume left and right unpaired point detection cells in addition to depth detection cells. These 3 types of cells cooperatively interact with each other depending on physical constraints (uniqueness, smoothness, occlusion) to estimate depth and determine which zones are unpaired. Moreover, since it is contradictory for monocularly visible zones to be visible in both eyes, we introduce mutual inhibition between left and right unpaired point detection cells. When input images including unpaired zones satisfy occlusion geometry, the model outputs the depth of the zones. The interesting finding is that when we input two different images to the eyes, the model shows an unstable output that alternates between interpretations of monocularly visible zones for the left and the right eyes, thereby reproducing binocular rivalry. Our results suggest that binocular rivalry is an erroneous output of a stereo mechanism that estimates the depth of half-occluded points. There are two general theories for what the rivals are in binocular rivalry: the two eyes, or representations of two different stimuli. We propose a new hypothesis that bridges these two: interocular inhibitions between representations of monocularly visible zones cause binocular rivalry. Unlike the traditional interocular theory, the level of the inhibitions here is after binocular convergence, thus open to a stimulus-specific mechanism.