Search Results

The application of n-dimensional transfer functions for feature segmentation has become increasingly popular in volume rendering. Recent work has focused on the utilization of higher order dimensional transfer functions incorporating spatial dimensions (x,y, and z) along with traditional feature space dimensions (value and value gradient). However, as the dimensionality increases, it becomes exceedingly difficult to abstract the transfer function into an intuitive and interactive workspace. In this work we focus on populating the traditional two-dimensional histogram with a set of derived metrics from the spatial (x, y and z) and feature space (value, value gradient, etc.) domain to create a set of abstract feature space transfer function domains. Current two-dimensional transfer function widgets typically consist of a two-dimensional histogram where each entry in the histogram represents the number of voxels that maps to that entry. In the case of an abstract transfer function design, the amount of spatial variance at that histogram coordinate is mapped instead, thereby relating additional information about the data abstraction in the projected space. Finally, a non-parametric kernel density estimation approach for feature space clustering is applied in the abstracted space, and the resultant transfer functions are discussed with respect to the space abstraction.