Efficient rendering of multiresolution meshes with guaranteed image quality

Output-sensitive rendering algorithms have recently generated a lot of inte rest. The number of triangles describing a scene was once smaller than the number of pixels. The model size is now significantly larger; the screen re solution remains fairly constant. In view-dependent use of multiresolution models for output sensitivity, one determines the lowest resolution necessa ry for object parts for good images. Some approaches use cones of normal ve ctors for illumination-dependent selective refining. The drawback: using on ly 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 betwe en the various levels of detail. This provides accurate lighting with a min imum number of triangles. In view-dependend use of multiresolution models f or output sensitivity, one determines the lowest possible resolution for ea ch object part that still gives good images.