Desiree - Detailed Surfaces for Interactive Rendering

Humans perceive their surroundings by the light that is reflected from surfaces. In particular, it is the fine detail observed on surfaces that provides clues about material properties like roughness, texture, temperature, etc. Similarly, in interactive computer graphics applications, the complexity of surface renderings is one of the first things that occur to an observer when judging how good or realistic the application looks. One key issue is to provide the same amount of visual detail as a real surface. This plays an important role in a number of interactive applications where the viewpoint and illumination change rapidly, including visual impact analysis, cultural heritage, design reviews, architecture and urban planning, driving and traffic simulation, engineering and computer games to name but a few. The aim of the Désirée project is to develop algorithms and data structures to efficiently acquire, store and display geometrically complex surfaces for such real-time applications. Adding surface detail has been a research topic since the early days of computer graphics, including texture mapping, displacement mapping and slice-based representations. However, none of these approaches can directly be used for the aforementioned applications due to insufficient image quality, non-interactivity or too high memory requirements. In addition, no tools are available to convert the complex models created by 3D artists into representations useful for fast and high-quality display. In Desiree, the main strategy will be to treat the rough object shape and the fine-scale details separately. We will research efficient data structures and algorithms that consider all aspects of this strategy, starting from the decomposition of a complex mesh into low- and high-detail components, efficient representations for the high- detail components, different mappings from the high-detail to the low-detail representation, and high-quality rendering in real time, including anti-aliasing issues and realistic illumination. For rendering, we will exploit recent programmable graphics hardware to develop output-sensitive display algorithms based on ray casting. We believe that this concept will allow us to achieve both, high image quality and real-time frame rates at the same time.

One core domain of the field of computer graphics is the generation of images from a given model. This is generally subsumed under the name Rendering and constitutes the domain of this project. As we developed fundamental techniques for this field, many of its application can profit from our results; the most important being industry and engineering (designing, virtual prototyping, urban planning), cultural heritage, entertainment (visual effects, computer games) and simulations (military, civil). A continuing trend in computer graphics is the demand for increased visual quality and enhanced realism in the generated images. On the one hand, this is enabled by the usage of more detailed models, but on the other hand, this development requires increased performance of the rendering system to enable a timely generation of the desired images. In this project, we investigated how to handle complex model data in interactive rendering, i.e., when the execution time of the image generation task has to be of the order of hundreds of milliseconds or less. As a result of our work, we provide methods to efficiently integrate fine details of geometrically complex surfaces into the rendering task. The first main body of our work is the use of diffusion curves to allow for exact geometric features in 2D and 3D graphics, irrespective of the how far the observer zooms in on the details. The second body is the use of exact mathematical computations to enable the rendering of fine details to raster images without suffering from the common quality degradations caused by aliasing artifacts. Both methodologies were developed with the recent advances in hardware technologies in mind and efficiently utilize graphics hardware to satisfy the requirement for interactive performance. As rendering is one part of the larger topics in computer graphics, we also explored related fields. In the work on second-order shape matching we supply a method to efficiently transfer information from one exemplar to similar objects in the modeling process, which provides the input for our rendering methods. For models that are represented as a huge cloud of points, which usually come from laser scans and depth imaging sensors, we provide a method to interactively generate a high-quality preview of the data. In summary, this project broke new ground in the highly efficient display of fine details of 2D and 3D scenes and provides fertile ground for further research, as well as new state-of-the art methods for applications.