What's next?

This is obviously, only a beginning; and even then, the beginning of only a single technique for the NPR representation of terrain. The present results are, indeed, promising. But more can be done. The algorithms presented should be further refined. Distance reinforcement for the curvature calculations should be added to the three curvature approximations currently lacking it (two-normal, three-vertex and edge-angle). The line-shading algorithm needs to allow lines to go in directions beyond the eight compass directions by weighting curvature directions (as the hatch shading has done.)

After even a small time interacting with terrain rendered in this manner, one simple principle, currently abrogated, becomes painfully clear: line density. The effectiveness of the line shader stands in direct proportion to the density of lines. If the camera draws too close to the image, the lines pull so far apart that they fail to communicate anything about the contour; too far and they merge into an incomprehensible mess. Some means of control needs to be applied to control line density. One approach is to use interpolated starting points. The current algorithm iterates through actual vertex locations as starting points for the lines. If the spaces between those vertices were subdivided, an arbitrary number of additional sets of lines could be created. They could be drawn or excluded, depending on the distance of the surface from the camera.

This, of course, suggests one of the most obvious directions in which research can, and should, continue: view-dependent modifications. Increasing and decreasing the number of lines drawn would be one such example. Others could include taking a heavily rendered image (like those in figures 3(e) or 4(d)) and actually lighting the vertices to help determine color, adding additional, more sophisticated contours (such as suggested contours), replacing simple line segments with other primitives, like polygons with brush stroke textures, etc. The inclusion of view-dependent concepts further implies research into accelerating these algorithms through the use of modern GPUs.

These images are all rendered in such a way to hint at pencil/ink drawings on paper. It is not strictly necessary to have the occluding surface in the hidden-line renderings to be "paper" colored. Combinations of textured surfaces and lines could yield very compelling images.

Mapmakers throughout history have understood the necessity and value of simplifying representations of terrain; extraneous data can be discarded and the viewer can be led to focus on specific features more effectively. With the advent of the internet, more data is publicly available than ever before. Much of this data can be tied together based on geographical location. Efficient and compelling non-photorealistic, interactive renderings of terrain are the first step to a new, compelling interface for accessing geo-sensitive data.

Researched and written by Sean Curtis with a great deal of help from Pete Shirley and Dave Gallup (August 16, 2004).