Image rendering by adaptive refinement

Abstract

This paper describes techniques for improving the performance of image rendering on personal workstations by using CPU cycles going idle while the user is examining a static image on the screen. In that spirit, we believe that a renderer's work is never done . Our goal is to convey the most information to the user as early as possible, with image quality constantly improving with time. We do this by first generating a crude image rapidly and then adaptively refining it where necessary as long as the user does not change viewing parameters. The renderer operates in a succession of phases, first displaying only vertices of polygons, next polygon edges, then flat shading polygons, then shadowing polygons, then Gouraud shading polygons, then Phong shading polygons, and finally anti-aliasing. Performance is enhanced by each phase using results from previous phases and trimming the amount of data needed by the next phase. In this way, only a fraction of the pixels in an image may be Phong shaded while the rest may be Gouraud or flat shaded. Similarly anti-aliasing is performed only on pixels around which there is significant color change. The system features fast response to user intervention, encourages user intervention at any moment, and makes useful the idle cycles in a personal computer.