Adaptive View-Dependent Tetrahedral Mesh Simplification


Erhan Okuyan
Computer Engineering Department
Bilkent University

Many practical application, such as MRI,CT machines or computational fluid dynamics applications produces volume data. Volume data are often represented using a tetrahedral mesh structure. Direct volume rendering tools uses tetrahedral mesh representations of volume data for visualization. However, the volume data generally are in quite high detail (up to 10 million tetrahedra), making the rendering quite slow. In order to have interactive rendering speed, volume data has to be simplified. There are many works that are proposed to change the detail level of the volume data. Edge collapse or tetrahedron collapse based approaches repetitively collapses the edge or tetrahedron, which will result in smallest quality loss in rendering. The inverse operation for these collapse operations are defined and they can be applied efficiently. Accordingly, these approaches produces a mesh representation scheme, whose detail level can be finely adjusted. However, these approaches are not view-dependent and the rendering quality suffers from it. Another approach, multi-resolution models, simplifies the whole mesh to a few level of details and then merge segments of volume from different levels of details. This approach is view-dependent, however it is quite restrictive. Our approach is taking the edge or tetrahedron collapse based approach and modifying it to have view dependent characteristics. In current approach if you want to go finer or coarser in the mesh there is only one operation for each choice. The operation sequence is implemented like a linked list structure, which you can go just two ways. However, if two collapse operations have no common affection areas, they can be executed interchangeably. In this work, we are tying to exploit that property. We are trying to convert the linked-list style operation sequence into a tree-style one. This approach allows changing the level of detail for different parts of the mesh differently. Accordingly the desired level of detail can be obtained in a view dependent manner, without having the restrictions of multi-resolution models.


DATE: 3 November, 2008, Monday@ 16:40