Self-organized figure-ground segmentation by multiple-cue integration