Image-Based Interactive Visualization of Large-Scale Data Sets

With the development of supercomputers, the parallel scale of numerical simulations is increasing exponentially. It is a time-consuming task to visualize the data sets of ever-increasing scale output in the simulations. It is difficult or even impossible to analyze the data sets smoothly, even on a visualization server of high performance. To achieve good interactivity and display effect, an efficient image-based approach is proposed in this paper, and a corresponding software system is also developed. Firstly, typical visual analysis patterns are summarized based on domain knowledge. These patterns can commendably depict inner physical characters in a simulated data field. Secondly, based on these patterns, a set of high-resolution images are pre-generated from large-scale time-varying data sets on a supercomputer. These images are rendered at various viewpoints around the data field and then organized according to the patterns and the time steps. Thirdly, we analyze the simulated data field interactively from these visualization images on a common computer. These visualization images can be observed from different directions by changing the viewpoint interactively. They can also be shown in proper resolution according to the size of the observing window. As a result, the entire time-varying evolution of physical characters can be interactively explored at different viewpoints in proper resolution. The proposed approach does not suffer from large scale of data sets and high computational complexity of visualization and thus significantly improves the visual analysis efficiency for large-scale data sets from the simulations.