high performance computing on graphics processing units: hgpu.org

Measuring the similarity between two streamlines is fundamental to many important flow data analysis and visualization tasks such as feature detection, pattern querying and streamline clustering. This dissertation presents a novel streamline similarity measure inspired by the bag-of-features concept from computer vision. Different from other streamline similarity measures, the proposed one considers both the distribution of and the distances among features along a streamline. The proposed measure is tested in two common tasks in vector field exploration: streamline similarity query and streamline clustering. Compared with a recent streamline similarity measure, the proposed measure allows users to see the interesting features more clearly in a complicated vector field. In addition to focusing on similar streamlines through streamline similarity query or clustering, users sometimes want to group and see similar features from different streamlines. For example, it is useful to find all the spirals contained in different streamlines and present them to users. To this end, this dissertation proposes to segment each streamline into different features. This problem has not been studied extensively in flow visualization. For instance, many flow feature extraction techniques segment streamline based on simple heuristics such as accumulative curvature or arc length, and, as a result, the segments they found usually do not directly correspond to complete flow features. This dissertation proposes a machine learning-based streamline segmentation algorithm to segment each streamline into distinct features. It is shown that the proposed method can locate interesting features (e.g., a spiral in a streamline) more accurately than some other flow feature extraction methods. Since streamlines are space curves, the proposed method also serves as a general curve segmentation method and may be applied in other fields such as computer vision. Besides flow visualization, a pedagogical visualization tool DTEvisual for teaching access control is also discussed in this dissertation. Domain Type Enforcement (DTE) is a powerful abstraction for teaching students about modern models of access control in operating systems. With DTEvisual, students have an environment for visualizing a DTE-based policy using graphs, visually modifying the policy, and animating the common DTE queries in real time. A user study of DTEvisual suggests that the tool is helpful for students to understand DTE.