Data fusion of inputs from fundamentally different imaging techniques requires the identification of a common subset to allow for registration and alignment. In this paper, we describe how to reduce the isosurface of a volumetric object representation to its exterior surface, as this is the equivalent amount of data an optical surface scan of the very same specimen provides. Based on this, the alignment accuracy is improved, since only the overlap of both inputs has to be considered. Our approach allows for a rigorous reduction below 1% of the original surface while preserving salient features and landmarks needed for further processing. The presented algorithm utilizes neighborhood queries from random points on an ellipsoid enclosing the specimen to identify data points in the mesh. Results for a real world object show a significant increase in alignment accuracy after reduction, compared to the alignment of the original representations via standard approaches.