Evaluating Qualitative Models of Shape Representation

Creating intelligent sketch-based educational software requires representing sketches the way a person would. Here, we evaluate several qualitative models of shape representation to determine which one best matches human judgments. The models differ in a) the way they segment an object’s contour into meaningful edges; and b) their level of detail for describing a shape. Our analysis indicates our improved edge-segmentation algorithm produces better results than an older algorithm, and a representation focusing on the edges in each shape best matches human judgments.