Efficient rendering of multiresolution meshes with guaranteed image quality

Output-sensitive rendering algorithms have recently generated a lot of interest. The number of triangles describing a scene was once smaller than the number of pixels. The model size is now significantly larger; the screen resolution remains fairly constant. In view-dependent use of multiresolution models for output sensitivity, one determines the lowest resolution necessary for object parts for good images. Some approaches use cones of normal vectors for illumination-dependent selective refining. The drawback: using only the normal cone for detection is a poor criterion in problematic cases, leading to too much time-consuming refinement. We describe a new approach for better estimation of normal deviations between the various levels of detail. This provides accurate lighting with a minimum number of triangles. In view-dependend use of multiresolution models for output sensitivity, one determines the lowest possible resolution for each object part that still gives good images.