In this dissertation, we show how we can use principles of saliency to enhance depiction, manage visual attention, and increase interactivity for 3D graphics and visualization. Current mesh saliency approaches are inspired by low-level human visual cues, but have not yet been validated. Our eye-tracking-based user study shows that the current computational model of mesh saliency can well approximate human eye movements. Artists, illustrators, photographers, and cinematographers have long used the principles of contrast and composition to guide visual attention. We present a visual-saliency-based operator to draw visual attention to selected regions of interest. We have observed that it is more successful at eliciting viewer attention than the traditional Gaussian enhancement operator for visualizing both volume datasets and 3D meshes.

Mesh saliency can be measured in various ways. The previous model of saliency computes saliency by identifying the uniqueness of curvature. Another way to identify uniqueness is to look for non-repeating structure in the middle of repeating structure. We have developed a system to detect repeating patterns in 3D point datasets. We introduce the idea of creating vertex and transformation streams that represent large point datasets via their interaction. This dramatically improves arithmetic intensity and addresses the input geometry bandwidth bottleneck for interactive 3D graphics applications.

Fast-previewing of time-varing datasets is important for the purpose of summarization and abstraction. We compute the salient frames in molecular dynamics simulations through the subspace analysis of the protein's residue orientations. We first compute an affinity matrix for each frame i of the simulation based on the similarity of the orientation of the protein's backbone residues. Eigenanalysis of the affinity matrix gives us the subspace that best represents the conformation of the current frame i. We use this subspace to represent the frames ahead and behind frame i. The more accurately we can use the subspace of frame i to represent its neighbors, the less salient it is.

Taken together, the tools and techniques developed in this dissertation are likely to provide the building blocks for the next generation visual analysis, reasoning, and discovery environments.