A graph-based approach to image segmentation uses an object’s bounding box to transform image pixels into a weighted graph. A shortest-path search then traces a closed contour, with reported tests demonstrating its use on vehicle images and other object classes.