Large Model Visualization : Techniques and Applications

The size of datasets in scientific computing is rapidly increasing. This increase is caused by a boost of processing power in the past years, which in turn was invested in an increase of the accuracy and the size of the models. A similar trend enabled a significant improvement of medical scanners; more than 1000 slices of a resolution of 512x512 can be generated by modern scanners in daily practice. Even in computer-aided engineering typical models eas-ily contain several million polygons. Unfortunately, the data complexity is growing faster than the rendering performance of modern computer systems. This is not only due to the slower growing graphics performance of the graphics subsystems, but in particular because of the significantly slower growing memory bandwidth for the transfer of the geometry and image data from the main memory to the graphics accelerator. Large model visualization addresses this growing divide between data complexity and rendering performance. Most methods focus on the reduction of the geometric or pixel complexity, and hence also the memory bandwidth requirements are reduced. In this dissertation, we discuss new approaches from three different research areas. All approaches target at the reduction of the processing complexity to achieve an interactive visualization of large datasets. In the second part, we introduce applications of the presented ap-proaches. Specifically, we introduce the new VIVENDI system for the interactive virtual endoscopy and other applications from mechanical engineering, scientific computing, and architecture.The size of datasets in scientific computing is rapidly increasing. This increase is caused by a boost of processing power in the past years, which in turn was invested in an increase of the accuracy and the size of the models. A similar trend enabled a significant improvement of medical scanners; more than 1000 slices of a resolution of 512x512 can be generated by modern scanners in daily practice. Even in computer-aided engineering typical models eas-ily contain several million polygons. Unfortunately, the data complexity is growing faster than the rendering performance of modern computer systems. This is not only due to the slower growing graphics performance of the graphics subsystems, but in particular because of the significantly slower growing memory bandwidth for the transfer of the geometry and image data from the main memory to the graphics accelerator. Large model visualization addresses this growing divide between data complexity and rendering performance. Most methods focus on the reduction of the geometric or pixel complexity, and hence also the memory bandwidth requirements are reduced. In this dissertation, we discuss new approaches from three different research areas. All approaches target at the reduction of the processing complexity to achieve an interactive visualization of large datasets. In the second part, we introduce applications of the presented ap-proaches. Specifically, we introduce the new VIVENDI system for the interactive virtual endoscopy and other applications from mechanical engineering, scientific computing, and architecture