Running the figure to the ground: Figure-ground segmentation during visual search