The primary research interest of my group is the development of algorithms and tools for fast and efficient generation as well as manipulation of high-quality, realistic imagery. In particular, we have been researching a number of challenging problems in the following sub-areas:
Rendering of realistic objects under general lighting requires the simulation of complex light transport phenomena like shadows, inter-reflections, caustics and sub-surface scattering. We proposed one of the first methods that could deal with all these phenoma using a technique called Precomputed Radiance Transfer. Roughly speaking, light transport is precomputed and projected into basis functions, which enables real-time evaluation of lighting under arbitrary illumination. Since this technique is rather efficient, it has been incorporated into DirectX and is used by a number of games, such as Halo 2. Precomputed radiance transfer techniques are inherently limited to static scenes, falling short of the ultimate goal of realistic, dynamic imagery. Recently, we have started to work on real-time global illumination techniques that enable the use of fully dynamic geometry, lights, and materials. We are investigating the use of approximate visibility, which has already led to very promising results. Further, we have been investigating another important aspect of real-time rendering: shadows. Shadowing techniques are commonly limited to the generation of hard shadows. Even then, anti-aliasing is a common artifact. We have developed a new mathematical formulation of shadowing, Convolution Shadow Maps, which enables us to efficiently render aliasing-free shadows as well as soft shadows.
Realistic renderings require not physically-based simulation of light transport but also realistic materials. To this end, we have worked on the acquisition and representation of realistic material properties, such as reflectance acquisition and texture transfer. Editing of complex material properties is another area, which we have recently addressed and where we have shown that common image editing operations transfer into the domain of material editing. We have also worked on real-time, realistic rendering of objects with complex material properties, such as subsurface scattering, glossy materials under environmental illumination, or hair.
More recently, we have started working in the area of computational photography, and more specifically in the area of high-dynamic-range imaging (HDRI). HDRI is a set of techniques that allows one to increase the dynamic range of traditional photographs, thus avoiding under- or overexposed areas. To this end a bracketed sequence of images is concatenated in one single HDR image. In order to display an HDR image, one then needs to compress the dynamic range again into the dynamic range of the display, which is commonly called tone-mapping. We have proposed a technique called Exposure Fusion, which directly creates a tone-mapped image without going through an HDR image. This technique was quickly adopted by a number of tools, e.g., PanoTools, due to its robustness (no parameter tweaking required). We have further worked on radiometrically calibrated HDR imaging (i.e., both in terms of color and luminance), which enables the use of a normal digital SLR camera instead of expensive measurement devices.