I. EXPERIMENT AND ANALYSIS The experimental environment in this paper is shown in table 1. We experimented our method with a model of a beetle, the rendering results are shown in fig. 3. Fig. 3(b) is the rendering result with constant line thickness, it appears somehow stiff. At the detail regions such as tentacles, lines are too thick and dense so that is not good for depicting the details. In fig. 3(c) and fig. 3(d), line thickness has more reasonable variation. It can be observed obviously that at the detail regions, our result gives thinner lines which is good for depicting details. At the dark regions such as the back of the beetle, our result gives thicker lines, which is good for emphasizing the shape. Though our result is not quite the same as Goodwin's, it finely meets the above 4 aesthetic rules. We experimented our method with a human model, fig. 4 shows the rendering results of its right arm. In fig. 4(c), we can see the line thickness is affected by depth, size relationship and brightness simultaneously. The palm is closer to the camera than the upper arm, so it has thicker lines. The back of the hand is flat while the fingers are thin, so lines at fingers are thinner than those at the back of the hand. Compared to the line drawing method using uniform line thickness, our method gives a more aesthetically pleasing and harmonious result. Since our method is based on triangle shell technique, it can't draw the lines (such as suggestive contours) except
silhouette outlines. In fig. 3(c), some detail lines at the beetle's legs didn't appear in fig. 3(d). But, in practice, we seldom use only lines in a stylized rendering application. When we shade the interior region properly, there will be enough visual cues for viewers to percept the shape. So, the lines except silhouette outlines are often unnecessary in many cases. The rendering speed of our method is shown in table 2. Since we have already baked the view-independent curvature information into a texture, we can read the data during the run-time process, instead of recalculating the information for