Theses Doctoral

Synthesis, Editing, and Rendering of Multiscale Textures

Han, Charles

The study of textures---images with repeated visual content---has produced a number of useful tools and algorithms for analysis, synthesis, editing, rendering, and a variety of other applications. However, the recent rapid growth in data storage and computational abilities has expanded the notion of what constitutes a texture. Modern textures can often outstrip traditional assumptions on input size by several orders of magnitude. Additionally, these multiscale textures typically contain features at not just one scale but rather across a wide range of scales, further violating existing assumptions. In order to meaningfully capture the large-scale features present in multiscale textures, we introduce a new example-based input representation, the exemplar graph. This representation enables allows us to efficiently define textures spanning a large--or possibly infinite--range of visual scales. We develop a hierarchical, parallelizable algorithm for performing texture synthesis from an input exemplar graph. In addition to automated generation, an increasingly important application of texture synthesis is in interactive tools for guiding texture design. This modality is especially important for multiscale textures, as they offer special perceptual challenges to artists. We examine algorithmic and engineering optimizations to enable real-time analysis and synthesis of multiscale textures, and explore potential implications for editing tools. Finally, we study the issue of display. To accurately view a large image at distance, some filtering operation must be performed. In many cases, such as traditional color images, the filtering operations are well-known. However, other texture representations, such as normal or displacement maps, present special difficulties for filtering. We treat the former case, presenting a principled analysis and algorithms for filtering and display of large normal maps.



  • thumnail for Han_columbia_0054D_10223.pdf Han_columbia_0054D_10223.pdf application/pdf 11.5 MB Download File

More About This Work

Academic Units
Computer Science
Thesis Advisors
Grinspun, Eitan
Ramamoorthi, Ravi
Ph.D., Columbia University
Published Here
May 17, 2011