CAREER: Efficient and Realistic Spatial/Temporal Appearance Details

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

0132970 Yu, Yizhou U of Ill - Champaign This project addresses the problem of recovering and synthetically generating realistic appearance details. Appearance details include color/reflectance variations, small geometric features on 3D surfaces, and the change of these high-resolution geometric and physical properties at different times. We call these details 3D and temporal textures in contrast to 2D textures which only have color variations in a 2D space. This project describes fundamental research in four classes of problems surrounding 3D and temporal textures: (1) the development of a unified framework for texture modeling considering 3D spatially and temporally varying details; (2) the development of novel algorithms that either recover high-quality appearance details from photographs and videos or generate them synthetically; (3) the development of efficient synthesis algorithms that can generate novel instances of 3D and temporal textures from given examples; (4) the investigation of efficient representation and compression schemes for 3D and temporal textures. This project will result in significant advances in 3D and temporal appearance modeling and synthesis, and therefore lead to overall qualitative improvement of the synthetic imagery created by graphics approaches. 3D and temporal textures complement geometric modeling and physical simulation. Addressing the proposed problem will greatly benefit these and other related research directions. Recent advances in physics-based simulation made it possible to generate synthetic imagery from solving light transport and motion equations. While these simulation-based techniques are quite expensive, the results generated from these techniques are often found to be too clean and smooth to look realistic enough. Incorporating appearance details into these methods would be highly desirable. Meanwhile, recent geometric reconstruction techniques from images and laser range scans have made it possible to acquire high-quality geometric models. Appearance modeling is becoming increasingly important in computer graphics applications where the efficient and realistic representation of details is essential. This work will be able to produce statistically correct, physically plausible appearance details in an efficient manner. In the film industry, high-quality hair models and skin surface details will dramatically shrink the gap between synthetic and real characters. In the game industry, real-time 3D and temporal textures will bring virtual environments a large step closer to reality. This holds true for a large number of applications including electronic commerce, TV and Web content production, human computer interaction, architectural and archeological walkthroughs. Altogether, the expected outcome of the project will be the development of important new techniques and results that provide real-world applications with useful solutions, cross-fertilization between computer graphics and other related areas (such as image processing and computer vision), and curricular innovations.