Out-of-core multiresolution techniques for graphics compression and volume visualization of large datasets

The rapid growth of the data size in recent years has made out-of-core techniques indispensable, where the datasets are too large to fit in main memory and we want to design new computational algorithms to reduce the I/O communications between main memory and disk. In this thesis we present our new out-of-core techniques for graphics compression and volume visualization of large datasets. Our focus is on multiresolution/level-of-detail (LOD) approaches, where we simplify the data and build a multiresolution hierarchy so that we can render the data at just the right level to achieve both high image quality and fast computing speed, in the out-of-core setting where the techniques work well for datasets larger than main memory. We develop out-of-core multiresolution approaches on three different types of datasets and computing tasks. First we focus on 3D triangle meshes, and develop a novel progressive lossless compression algorithm that supports selective decompression. We then extend our study to time-varying regular-grid volume rendering. We explore the temporal and spatial coherences of the dataset to speed up the volume rendering, by developing a new tree structure that supports a high re-use rate of the sub-volumes, as well as devising the corresponding I/O efficient algorithms that are novel and highly non-trivial. Finally, we develop a volume rendering algorithm for irregular-grid volume data represented as tetrahedral meshes. We devise a novel out-of-core simplification and level-of-detail (LOD) volume rendering algorithm where the underlying LOD mesh is guaranteed to be crack-free, namely, any neighboring sub-volumes in the LOD mesh have consistent boundaries, and all the cells in the LOD mesh are fold-over free (i.e., do not have negative volumes). Our technique supports selective refinement LODs, in addition to the basic uniform LODs. The proposed scalar-value range and view-dependent selection queries for selective refinement are especially effective in producing images of the highest quality with a much faster rendering speed. We present experimental results which show the efficacy of these new out-of-core techniques.