Segmentation, modeling, and visualization of complex features in large volume data

In many fields of study such as medical imaging and nondestructive testing (NDT), scientists generate very high resolution volume data to detect and analyze fine and complex features of interest. While the high resolution of the data provides more accurate information about the subject of study, it also presents great challenges to the associated data analysis tasks. This dissertation addresses these challenges to enable interactive segmentation, modeling, and visualization of complex features in large volume data using a single PC.; I have developed a two-step process which allows the user to balance between the interactivity and accuracy requirements for effectively studying volumetric features. In the first step, three techniques have been designed to interactively label the feature of interest including a region growing based technique, a three-level graph based approach, and a multi-scale morphological data exploration method. All the algorithms make use of the statistical characteristics of the feature of interest, as well as its local structural information. The second step receives the segmentation result from the first step and automatically constructs a feature boundary at sub-voxel accuracy. Three representations including points, volume and surfaces can be derived by taking into account the boundary's uncertainty information using a boundary model.