Abstract
Until now computer graphic researchers have tried to solve visualization problems introduced by the size of meshes. Modern tools produce large models and hardware is not able to render them in full resolution. For example, the digital Michelangelo project extracted a model with more than one billion polygons. One can notice hardware has become more and more powerful but meshes have also become more and more complex. To solve this issue, people have worked on many solutions. We can find solutions based on space subdivision, or based on visibility of objects like the use of a Z-buffer. But in 1976, Clark [Clark 1976] introduces the level of detail concept (LOD). The principle of LOD is the construction of several versions of the same 3D model at different resolutions. This is achieved by removing some object features. Luebke provides in [Luebke 1997] a very complete survey of LOD algorithms. The main issue with the simplification is that the mesh does not preserve appearance of the original mesh. Indeed, important features tend to disappear. For example, with the Quadric Error Metrics (QEM) algorithms and the cow mesh, the tail, horn and other characteristic points merge with the mesh at a low resolution. Our approach allows the simplified mesh to preserve important details.