Diploma and Master Theses (authored and supervised):
"Coloring Meshes of Archaeological Datasets";
Supervisor: M. Wimmer;
Institut für Computergraphik und Algorithmen,
Archaeological monuments are nowadays preserved for future generations by means of digitization. To this end, laser scanners in conjunction with registered cameras are used to gather both the geometric and the color information. The geometry is often triangulated to allow a continuous mapping of the photos onto the geometry. The color mapping for high-quality reconstructions of parts of the models is not without problems. In practice, the photos overlap. Now, assuming that a particular triangle receives color information from just one photo, there is a huge number of possibilities to map the photos onto the triangles. This labeling problem is already covered in literature. There are also approaches for the leveling of the remaining seams that arise because of the different lighting situations during the exposure of the photos. In this thesis, we improve common labeling approaches by the introduction of an accurate geometry-based occlusion detection. An octree is used to quickly filter out parts of the model that do not come into consideration for an occlusion anyway. The occlusion detection prevents texturing of parts of the model with image material that does not contain the expected region, but the colors of an occluder.
Further, a proposed approach for seam leveling on meshes is improved by the introduction of a new term into the least squares problem that corresponds to the proposed leveling approach. This new term penalizes big leveling function values and helps to keep the leveled color values in the valid range. For better filtering results, we improve the proposed calculation of a 1-pixel wide boundary around the leveled patches by the introduction of outline normals for a user-defined scale of the patches.
For easier manual editing of remaining artifacts in the photos, we introduce an application for the generation of alpha masks that indicate regions of the photos that are used for texturing of the 3D model.
For the high-performance visualization of 3D models with a huge amount of textures, we make use of virtual texturing. We present an application that generates the needed data structures atlas and tile store in significantly less time than existing scripts.
Finally, we show how all the mentioned functionalities are integrated into a visualization application that can support a graphic artist in the post-processing of a digitized archaeological monument.
Electronic version of the publication:
Created from the Publication Database of the Vienna University of Technology.