Speaker: Michael Birsak
In this paper we propose a framework for out-of-core real-time rendering of high-quality textured archaeological data-sets. Our input is a triangle mesh and a set of calibrated and registered photographs. Our system performs the actual mapping of the photos to the mesh for high-quality reconstructions, which is a task referred to as the labeling problem. Another problem of such mappings are seams that arise on junctions between triangles that contain information from different photos. These are are approached with blending methods, referred to as leveling. We address both problems and introduce a novel labeling approach based on occlusion detection using depth maps that prevents texturing of parts of the model with images that do not contain the expected region. Moreover, we propose an improved approach for seam-leveling that penalizes too large values and helps to keep the resulting colors in a valid range. For high-performance visualization of the 3D models with a huge amount of textures, we make use of virtual texturing, and present an application that generates the needed texture atlas in significantly less time than existing scripts. Finally, we show how the mentioned components are integrated into a visualization application for digitized archaeological sites.