What computer vision technique is utilized by an API when it returns an annotated image with bounding box coordinates for detected objects?