Abstract
Incompatible images presented to the two eyes lead to perceptual oscillations in which one image at a time is visible. Early models portrayed this binocular rivalry as involving reciprocal inhibition between monocular signals from each eye, occurring at an early stage prior to binocular mixing. However, subsequent psychophysical experiments found conditions where rivalry could occur between the two stimuli (dissociated from eye-of-origin) rather than between the two eyes, suggesting rivalry mechanisms at a higher, more abstract level of representation. Furthermore, neurophysiological recordings in monkeys found the strongest rivalry correlate in inferotemporal cortex, a high-level visual area involved in object recognition with predominantly binocular cells. These findings led to the formulation of hierarchical models of rivalry, in which previous models incorporating reciprocal inhibition between monocular image representations were augmented by an additional stage involving inhibition between higher-level, binocular representations of the two stimuli. An unresolved issue for current models is how can the separate identities of the two images be maintained after binocular mixing in order for rivalry to be possible at higher levels? Here we show that after left and right images are binocularly mixed, they can be unmixed at any subsequent stage using a signal processing algorithm, non-negative matrix factorization (NMF), previously proposed for parsing object parts during object recognition. Left and right images were combined in various proportions to form a set of different binocular mixtures, consistent with physiological observations of binocular ocular dominance classes. This set of binocular mixtures was then fed into the NMF algorithm, which regenerated the original left and right images from them. I suggest that binocular rivalry occurring at higher visual levels may be related to mechanisms of shape representation during object recognition rather than mechanisms of stereopsis.