Abstract
Studies using tasks such as visual search, divided attention, and multiple object tracking all suggest that the visual system quickly parses the retinal image into discrete objects using a set of inflexible rules. Here, we present rapid number estimation as a new paradigm for investigating how these rules operate quickly over the entire visual array. Subjects were asked to determine which of two briefly presented (450ms) displays contained more squares. Each display could contain between 6 to 52 squares, and the difference in number between displays was always around 30%. Each square was attached by its vertices to 4 parallel lines, and some squares shared these lines with other squares, making them appear as parts of a single object (opposite faces within an elongated “necker cube”). By varying the percentage of squares that formed connected pairs, we tested whether observers could resist seeing the connected squares as single objects, and consequently counting them as a single object, even though they were instructed to count only squares (and to completely ignore the lines). Critically, manipulating the percentage of connected squares did not alter the density or area subtended by the objects. We found that as the percentage of connected squares increased, observers increasingly underestimated the number of squares in that display, as if the connected pairs were often counted as 1 object instead of 2. In the most drastic case, when all the squares in the display containing more squares were connected, subjects only correctly identified the connected display as having more squares than the unconnected display 50% of the time, compared to 75% when comparing two unconnected displays. These findings suggest that rapid number estimation operates over visual objects which are defined by rules such as connectedness. This new paradigm provides a robust new method for further exploring the rules used by the visual system to quickly segment displays into objects.