Thesis (Ph. D.)--University of Texas at Austin, 2008. / Vita. Includes bibliographical references.

M.Sc. / Real-time terrain rendering (RTTR) is an exciting eld in computer graphics. The algorithms and techniques developed in this domain allow immersive virtual environments to be created for interactive applications. Many di culties are encountered in this eld of research, including acquiring the data to model virtual worlds, handling huge amounts of geometry, and texturing landscapes that appear to go on forever. RTTR has been widely studied, and powerful methodologies have been developed to overcome many of these obstacles. Complex natural terrain features such as detailed vertical surfaces, overhangs and caves, however, are not easily supported by the majority of existing algorithms. It becomes di cult to add such detail to a landscape. Existing techniques are incredibly e cient at rendering elevation data, where for any given position on a 2D horizontal plane we have exactly 1 altitude value. In this case we have a many-to-1 mapping between 2D position and altitude, as many 2D coordinates may map to 1 altitude value but any single 2D coordinate maps to 1 and only 1 altitude. In order to support the features mentioned above we need to allow for a many-to-many mapping. As an example, with a cave feature for a given 2D coordinate we would have elevation values for the oor, the roof and the outer ground. In this dissertation we build upon established techniques to allow for this manyto- many mapping, and thereby add support for complex terrain features. The many-to-many mapping is made possible by making use of geometry images in place of height-maps. Another common problem with existing RTTR algorithms is texture distortion. Texturing is an inexpensive means of adding detail to rendered terrain. Many existing technique map texture coordinates in 2D, leading to distortion on steep surfaces. Our research attempts to reduce texture distortion in such situations by allowing a more even spread of texture coordinates. Geometry images make this possible as they allow for a more even distribution of sample positions. Additionally we devise a novel means of blending tiled texture that enhances the important features of the individual textures. Fully sampled terrain employs a single global texture that covers the entire landscape. This technique provides great detail, but requires a huge volume of data. Tiled texturing requires comparatively little data, but su ers from disturbing regular patterns. We seek to reduce the gap between tiled textures and fully sampled textures. In particular, we aim at reducing the regularity of tiled textures by changing the blending function. In summary, the goal of this research is twofold. Firstly we aim to support complex natural terrain features|speci cally detailed vertical surfaces, over-hangs and caves. Secondly we wish to improve terrain texturing by reducing texture distortion, and by blending tiled texture together in a manner that appears more natural. We have developed a level of detail algorithm which operates on geometry images, and a new texture blending technique to support these goals.

Modern rendering systems require fast and efficient acceleration structures in order to compute visibility in real time. I present several novel data structures and algorithms for computing visibility with high performance. In particular, I present two algorithms for improving heuristic based acceleration structure build. These algorithms, when used in a demand driven way, have been shown to improve build performance by up to two orders of magnitude. Additionally, I introduce ray tracing in perspective transformed space. I demonstrate that ray tracing in this space can significantly improve visibility performance for near-common origin rays such as eye and shadow rays. I use these data structures and algorithms to support a key hypothesis of this dissertation: “There is no silver bullet for solving the visibility problem; many different acceleration structures will be required to achieve the highest performance.” Specialized acceleration structures provide significantly better performance than generic ones and building many specialized structures requires high performance build techniques. Additionally, I present an optimization-based taxonomy for classifying acceleration structures and algorithms in order to identify which optimizations provide the largest improvement in performance. This taxonomy also provides context for the algorithms I present. Finally, I present several novel cost metrics (and a correction to an existing cost metric) to improve visibility performance when using metric based acceleration structures. / text

A fundamental operation in computer graphics is to determine for a given point and direction in a scene, which geometric surface is nearest this point from this direction and thus visible. Conceptually, the point and direction define a "ray". Z-buffer hardware can compute surface visibility for a set of rays with a common origin (i.e. eye point) and a regular pattern of directions in real-time. However, this hardware is much less efficient at performing other visibility computations such as those required to accurately render shadows. A more flexible solution to the visible surface problem is needed. This work introduces the irregular Z-buffer algorithm, which efficiently solves the visible surface problem for rays with a common origin and arbitrary directions. In addition, we identify several changes to classical graphics architectures needed for hardware acceleration of this algorithm. Though these modifications are incremental in nature (i.e. no new functional units are introduced), we show that they enable significant new capability. In tandem with the irregular Z-buffer algorithm, a GPU with these changes has applications in: shadow rendering, indirect illumination, frameless rendering, adaptive anti-aliasing, adaptive textures, and jittered sampling. We explore the performance of hard and soft shadow rendering in particular, by way of a detailed hardware simulator. / text

Thesis (Ph. D.)--University of Texas at Austin, 2008. / Vita. Includes bibliographical references and index.

A seamless blending of the real and virtual worlds is key to increased immersion and improved user experiences for augmented reality (AR). Photorealistic and non-photorealistic rendering (NPR) are two ways to achieve this goal. Non-photorealistic rendering creates an abstract and stylized version of both the real and virtual world, making them indistinguishable. This could be particularly useful in some applications (e.g., AR/VR aided machine repair, or for virtual medical surgery) or for certain AR games with artistic stylization. Achieving temporal coherence is a key challenge for all NPR algorithms. Rendered results are temporally coherent when each frame smoothly and seamlessly transitions to the next one without visual flickering or artifacts that distract the eye from perceived smoothness. NPR algorithms with coherence are interesting in both general computer graphics and AR/VR areas. Rendering stylized AR without coherence processing causes the final results to be visually distracting. While various NPR algorithms with coherence support have been proposed in general graphics community for video processing, many of these algorithms require thorough analysis of all frames of the input video and cannot be directly applied to real-time AR applications. We have investigated existing NPR algorithms with coherence in both general graphics and AR/VR areas. These algorithms are divided into two categories: Model Space and Image Space. We present several NPR algorithms with coherence for AR: a watercolor inspired NPR algorithm, a painterly rendering algorithm, and NPR algorithms in the model space that can support several styling effects.

Thesis (M. S.)--Electrical and Computer Engineering, Georgia Institute of Technology, 2009. / Committee Chair: Jayant, Nikil; Committee Member: Altunbasak, Yucel; Committee Member: Sivakumar, Raghupathy. Part of the SMARTech Electronic Thesis and Dissertation Collection.

Transmission of medical images, that mandate lossless transmission of content over WLANs, presents a great challenge. The large size of these images coupled with the low acceptance of traditional image compression techniques within the medical community compounds the problem even more. These factors are of enormous significance in a hospital setting in the context of real-time image collaboration. However, recent advances in medical image compression techniques such as diagnostically lossless compression methodology, has made the solution to this difficult problem feasible. The growing popularity of high speed wireless LAN in enterprise applications and the introduction of the new 802.11n draft standard have made this problem pertinent. The thesis makes recommendations on the degree of compression to be performed for specific instances of image communication applications based on the image size and the underlying network devices and their topology. During our analysis, it was found that for most cases, only a portion of the image; typically the region of interest of the image will be able to meet the time deadline requirement. This dictates a need for adaptive method for maximizing the percentage of the image delivered to the receiver within the deadline. The problem of maximizing delivery of regions of interest of image data within the deadline has been effectively modeled as a multi-commodity flow problem in this work. Though this model provides an optimal solution to the problem, it is NP hard in computational complexity and hence cannot be implemented in dynamic networks. An approximation algorithm that uses greedy approach to flow allocation is proposed to cater to the connection requests in real time. While implementing integer programming model is not feasible due to time constraints, the heuristic can be used to provide a near-optimal solution for the problem of maximizing the reliable delivery of regions of interest of medical images within delay deadlines. This scenario may typically be expected when new connection requests are placed after the initial flow allocations have been made.