Date of Award
Doctor of Philosophy (PhD)
Warren N. Waggenspack, Jr
There are numerous techniques available for fitting a surface to any supplied data set. The feature-based modeling technique takes advantage of the known, geometric shape of the data by deforming a model having this generic shape to approximate the data. The model is constructed as a rational B-spline surface with characteristic features superimposed on its definition. The first step in the fitting process is to align the model with a data set using the center of mass, principal axes and/or landmarks. Using this initial orientation, the position, rotation and scale parameters are optimized using a Newton-type optimization of a least squares cost function. Once aligned, features embedded within the model, corresponding to pertinent characteristics of the shape, are used to improve the fit of the model to the data. Finally, the control vertex weights and positions of the rational B-spline model are optimized to approximate the data to within a specified tolerance. Since the characteristic features are defined within the model a creation, important measures are easily extracted from a data set, once fit. The feature-based modeling approach is demonstrated in two-dimensions by the fitting of five facial, silhouette profiles and in three-dimensions by the fitting of eleven human foot scans. The algorithm is tested for sensitivity to data distribution and structure and the extracted measures are tested for repeatability and accuracy. Limitations within the current implementation, future work and potential applications are also provided.
Dobson, Gregory Todd, "Feature-Based Models for Three-Dimensional Data Fitting." (1997). LSU Historical Dissertations and Theses. 6562.