Abstract
Theories of visual cognition have postulated a processing stage where elementary features are linked together to form more complex representations termed “object files” or “proto-objects”. The neural basis of the linking is not known. How is the representation of a square different from the representation of four lines? One hint comes from the observation of border ownership (BOS) selectivity in monkey visual cortex. About half of the neurons in area V2 are selective for which side of a border in the receptive field is the figure and which side is the ground. The left-hand side of a square, for example, produces high firing rates in neurons of figure-right preference and low firing rates in neurons of figure-left preference. These neurons combine information from various figure-ground cues, including stereoscopic depth, occlusion features and global shape. Do these neurons just integrate figure-ground cues, or do they reflect the formation of proto-object representations? One important characteristic of visual objects is continuity. The system can identify given objects across a sequence of changing images. If BOS signals reflect object-related coding, they should show this continuity. But if they merely represent the figure-ground cues, they should change whenever the cues change. To answer this question we devised stimuli in which the figure-ground cues reverse, while the objects remain the same. During an initial motion phase, the figure-ground cues indicate one side of ownership. At the end of the motion, the static configuration gives cues indicating the opposite side of ownership. We find that the initial BOS assignment persists for seconds despite the change in the figure-ground cues of the stimulus, indicating that the continuity of the objects dominates the neural response. Thus BOS signals reflect the emergence of proto-object representations.