Triangle Mesh Compression For Acceleration Rendering

It is often required a huge amount of triangles to represent complex 3D objects or scene for increasingly demanding of high rendering quality. Explosive growth in the complexity of triangle mesh becomes bottleneck for all kinds of graphics applications. An effective solution to this problem is to use a compressed mesh representation. However, most state-of-the-art compression algorithms compress an entire triangle mesh into a whole encoded mesh. The encoded mesh can't be partly accessed, which prevents triangle mesh compression from being effectively compatible with acceleration rendering techniques, such as PC-cluster parallel rendering, view-dependent out-of-core rendering, triangle strip rendering and cache friendly triangle strip rendering. In this paper, we study triangle mesh compression methods which are compatible with those acceleration rendering techniques. This thesis contributes to triangle mesh compression in the following aspects:Firstly, we propose a mesh compression scheme PRMC (Parallel Rendering based Mesh Compression). It supplies an encoded mesh which can be partly accessed in PC-cluster rendering and view-dependent out-of-core rendering. It includes three key technologies:1. We propose a high quality mesh segmentation algorithm. It produces submeshes with balance primitives. The triangles within a submesh have similar normals. The distances beween every vertex and its 1-ring neighbors within a submesh are similar. Based on it, PC-cluster parallel rendering system can calculate load balance effectively.2. We propose a compression algorithm based on topology preservation of submesh. The encoded submeshes can not only keep mesh topology information unchanged, but also be partly accessed. So every encoded submesh serves as a sorting primitive in parallel rendering.3. We extend the data type of scene graph (tree) of parallel rendering system. Secondly, we propose a single triangle strip generation and compression algorithm. A concentric circle is used to control triangle sequence in single triangle strip generation. Compared with other methods, it leads to less triangle splitting. When this algorithm is applied to view-coherent submesh, the normal locality and spatial locality of strips can be controlled. Moreover, it needs less code to represent an encoded single triangle strip.At last, we propose a hybird vertex cache friendly triangle stripification algorithm which based on the cache-aware algorithm. Combining the features of cache-oblivious algorithm, this method achieves a better tradeoff between cache-aware and cache-oblivious cache friendly triangle stripification algorithms.Experiental results showed that the compression methods proposed in this paper can be compatible better with acceleration rendering techniques.