Clouds: A model for synergistic image segmentation

Image segmentation consists of recognizing the object in the image and precisely delineating its spatial extent. We present a model, called clouds, that exploits the synergism which commonly exists between recognition and delineation for more effective segmentation. The model can reduce user's intervention to simple corrections or even eliminate it altogether, achieving high accuracy. We evaluate the method in the task of 3D MR image segmentation of the brain in isolating automatically: brain without medulla and spinal cord; just the cerebellum; and the brain hemispheres without medulla, spinal cord, and cerebellum. These structures are connected in several parts, which poses a serious challenge for simplistic segmentation strategies. The entire process takes a few seconds on modern PCs and provides accurate results. The applications for clouds go beyond medical imaging, opening new vistas in a variety of areas served by segmentation.