Compression of Three-Dimensional Models Through Vector Quantization

Tao Meng,P.H. Chou

2002-01-01

Abstract:The size and complexity of 3D models used in interactive computer graphics applications are rising rapidly, fueled by the proliferation of high-performance graphics processors and the ever-increasing demand for realism in graphics. Rendering detailed 3D models requires the transfer and processing of large amounts of data typically in the form of triangle meshes. Data compression is essential for the support of polygon transfer rates demanded by next-generation graphics processors. In this dissertation, we present an effective compression scheme for 3D triangle data based on predictive vector quantization. First, we introduce a lossless mesh connectivity coding algorithm based on a breadth-first traversal order that considerably reduces redundant vertex transmissions. The high locality of the resulting vertex sequence lends to effective predictive vector quantization of the vertex data. We explore methods using unstructured codebooks as well as a product code pyramid vector quantizer. Compression is not the only benefit offered by vector quantization; we also demonstrate how the technique can be used for complexity reduction by accelerating the computation of vertex transformations. Consequently, an encoded set of vertices can be both decoded and transformed in approximately 60 percent of the time required by a conventional method without compression. Using a similar approach, we extend our vector quantization method to two non-polygonal representations: surface light fields and volumetric data, for which we present an accelerated splat-based volume rendering algorithm that exploits the complexity-reducing properties of vector quantization. VQ is demonstrated to be an effective technique for dealing with the growing complexity of 3D datasets.