Efficient Representation of Captured Geometry and Reflectance

Tobias Nöll

PhD-Thesis TU Kaiserslautern 9/2015.


Having a digital representation of physical objects is desirable whenever realistic images of these objects need to be synthesized. This is the case for a large variety of use-cases including cultural heritage preservation (conservation and digital exhibition), e-Commerce (advertisment) and the production of movies (special effects). A digital representation should be compact, yet powerful enough to accurately resemble every single aspect defining the appearance of the physical counterpart. This visual appearance is mainly determined by the overall shape of the object and the interactions of its surface materials with incident light. We refer to the first proberty as geometry and to latter as reflectance. The work presented in this thesis is driven by the recent technical development of acquisition devices which enable synchronous aquiisition of object geometry and reflectance properties. We propose a novel methodology for the processing of raw data aquired by such devices that yields a compact, digital object under arbitrary, user-defined viewing and lightning conditions. The particulary novel aspect of our work is that we set great value upon the robustness and practicability of the developted concepts. We explicitly assume input with characteristics that are usually present in real measured data, i.e. massive, irregulary sampled and outlier afficted raw data.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence