Accurate and reliable extraction of surfaces from image data using a multi-dimensional uncertainty model.

Surface extraction is an important step in the image processing pipeline to estimate the size and shape of an object. Unfortunately, state of the art surface extraction algorithms form a straight forward extraction based on a pre-defined value that can lead to surfaces, that are not accurate. Furthermore, most isosurface extraction algorithms lack the ability to communicate uncertainty originating from the image data. This can lead to a rejection of such algorithms in many applications. To solve this problem, we propose a methodology to extract and optimize surfaces from image data based on a defined uncertainty model. To identify optimal parameters, the presented method defines a parameter space that is evaluated and rates each extraction run based on the remaining surface uncertainty. The resulting surfaces can be explored intuitively in an interactive framework. We applied our methodology to a variety of datasets to demonstrate the quality of the resulting surfaces.