Octree-Based Geometry Compression

Geometry compression of 3D models is a hot research area in current computer graphics. With the rapid development of computer technology and its applications, the size and the complexity of 3D models increase dramatically. This brings difficulties to store and transport 3D models over networks with limited bandwidth. Thus, it is necessary to develop bit efficient representations and transmitting efficient representations for 3D models. Compression of 3D models consists of geometry compression and connectivity compression. In general, the geometry portion of the content always dominates the compressed code so that many methods have been presented for efficient coding of geometric data to achieve high compression efficiency. In this thesis, a survey is first given on the techniques for geometry compression, and then a new progressive geometry encoder is proposed, which can achieve higher compression efficiency than the best geometry compression method known now in the field of model compression.A survey of geometry compression is given in the second chapter. It firstly introduces the basic pipeline for geometry compression and several principles for classifying geometry compression methods, and then divides the methods into two categories to discuss in details, the basic methods and the methods based on model transformation. Finally, according to our understanding of geometry compression, a summary and prospect for geometry compression are discussed.In the third chapter, we propose a new progressive geometry encoder. It is based on the adaptive octree to managing the vertices of the 3D model. The octree is constructed by recursively partitioning the cells containing vertices until the finest cells. By traversing the adaptive octree recursively with width-first, three bit streams are generated to implicitly represent vertices' coordinates. The three streams indicate if a cell contains vertices, if a cell contains only one vertex, and the location of the child_cell containing the vertex when a cell contains only one vertex respectively.Finally, the three bit streams are compressed respectively in arithmetic coders. In comparison with advanced similar progressive geometry coders, the new method can produce shorter bit streams. This provides a good basis for studying more efficient geometry coders. Our method can deal with all kinds of 3D geometric models, and can be easily extended to treat the models of higher dimensions and out-of-core models.