Generating perceptually more accurate images has been gaining popularity in computer graphics. However, such accuracy still demands expensive algorithms. As a result, some researchers have concentrated on approximation algorithms. This paper presents one such algorithm to generate images with realistic shadows. Interactive computer graphics systems commonly assume that the light source is at a single point (or in a single direction) when they generate shadows. The resulting shadows have sharp, or “hard” boundaries. Shadows from real light sources, on the other hand, do not have clearly visible boundaries, or in other words, are “soft.”
The efficiency of the proposed algorithm is due to the fact that the shadowed object must lie on a plane and that the generated shadows are simply a heuristic expansion of the boundaries (silhouette) of hard shadows into softer looking boundaries. This expansion is based on the distance of the shadowing object from the shadowed plane. The paper describes one way to combine the expansions when multiple shadowers exist. Furthermore, graphics display hardware is efficiently used to perform this combination. The resulting algorithm should be faster than the currently known schemes. Unfortunately, the paper presents neither the analysis of the speedup, nor a formal study of the effect of the expansion. It would be interesting to learn the characteristics of the errors introduced by the approximation, and to quantify those errors.
The paper is well written, even if the presentation is rather verbose.