A few words about my approach.

Features such as silhouette edges, ridges, valleys and intersections of geometric primitives are useful in NPR i.e. technical illustrations and artistic effects. NPR is ideal for many graphics programs: from course projects that demonstrate 3D phenomenon (e.g. collision detection) and interactive visualization to serious technical illustrations. However, we still see many programmers using traditional APIs (which can produce geometrically correct but photometrically terrible images) even when photorealism is not the goal. I believe that a wide-spead use of even simple NPR techniques is lacking due to the following reasons:

• Data structures do not support explicit rendering of edges, one needs to write significant code to identify edges, test whether they are silhouette/sharp and use two or more passes of rendering.
• Standard API's (due to hardware limitations) do not support operations that work on two or more polygons at a time. Finding silhouettes/sharp edges involves global processing of neighboring polygons. There are (were) no 'single-polygon-at-a-time' processing algorithms to identify silhouettes/sharp edges.

glEnable(GL_SILHOUETTE_EDGES); 
glEnable(GL_SHARP_EDGES); 
   render_object_triangle_soup();
glDisable(GL_SILHOUETTE_EDGES); 
glDisable(GL_SHARP_EDGES);