Karl vom Berge Katholieke Universiteit Leuven

Contact: Computer Graphics Research Group

Ph.D. Thesis, Katholieke Universiteit Leuven

Abstract

In the domain of computer graphics, more specifically photo realistic physically based image synthesis, realistic images are produced from a detailed scene model using physically based light transport simulations. Advances in light transport simulation methods have reached a point where all possible light paths in such a scene can be computed. Therefore, the accuracy and realism of synthesized images is especially dependent on the quality and accuracy of the scene model. An important aspect of scene models is the specification of reflection properties for all objects in the scene. In computer graphics, reflection from surfaces is typically described by high dimensional reflectance functions. For simple materials such as plastics, metals and glas, analytical models are very successful in describing their light scattering behaviour. For reflection of complex materials, like heterogeneous subsurface scattering and scattering from complex meso structures or organic materials, the formulation of analytical models is a very difficult task. An alternative approach to the specification of light scattering properties by analytical modelling is the measurement of reflectance functions from real-world surfaces or synthetically modelled meso structures. In this dissertation, the measurement of three classes of complex reflection functions – the bidirectional surface reflection distribution function (BRDF), the bidirectional texture distribution function (BTF) and the bidirectional surface scattering reflection distribution function (BSSRDF) – is investigated. In a first part of the dissertation, an acquisition device – a gonioreflectometer – for multispectral BRDF measurement of organic material is presented. Specific difficulties are discussed that arise when measuring organic materials such as plant leaves, and which do not occur when measuring inorganic materials. Functional requirements for the gonioreflectometer are identified and the construction of the device is detailled. Measurements of beech leaves are conducted, compared to existing reflection data, and used in the synthesis of realistic images of trees. In a second part of the dissertation, BTFs are measured from synthetically modelled meso structures. These measurements result in huge amounts of data which can not be used as such during rendering. A novel compact representation for the data is presented that is based on matrix factorization, which unifies previous approaches. Using the compact representation, it is shown that rendering of BTFs at interactive rates is possible using programmable graphics hardware. In the last part of the dissertation, a novel aquisition setup is presented for the measurement of heterogeneous BSSRDFs. A novel compact representation is presented, that is based on non-negative matrix factorisation, which is tailored to the specific structure of the measured data. This representation is spatially compact and can be evaluated efficiently during rendering. The technique is tested by measuring material models of several real-world objects which can be applied to novel geometry to produce realistic visualizations.

---