Abstract
We present a method for comprehensible non-photorealistic rendering of 3D free-form objects. The objects are rendered using a small amount of lines such as silhouettes or hatching strokes. The lines are carefully placed to ensure the user would understand the structure of the rendered object correctly. The user's "mental image" of the object is estimated using simplified methods of computer vision, allowing 3D shape reconstruction from the rendered image. The difference between the original model and the estimated "mental image" is used as a feedback for refining the line placement. The process stops when the difference is sufficiently low, which means the users would understand the rendering sufficiently well despite its simplicity.