Studies in Texture Generation of Weathered Appearance from Captured Data

Materials often change in appearance with time as a result of their interaction with the world around them. The rate and degree of weathering on an object are dependent on the shape of the object, its exposure to the surrounding environment, and the properties of the material of which the object is composed. Modeling temporal variations in material appearance is crucially important to the computer-generation of photo-realistic images. To date, most approaches for generating the appearance of weathered materials have been simulation-based. In this thesis, we describe a series of studies aimed at generating textures of weathered materials from captured data. First, to facilitate the synthesis of uniform weathered textures from natural images, we introduce a fully automatic pipeline to identify dominant texture samples based on a diffusion distance manifold. We define the characteristics of dominant texture and outliers that allow us to efficiently identify dominant texture in feature space. We also describe a psychophysical experiment to quantitatively validate our selection of the diffusion distance in the data manifold recovery. Second, we describe an appearance model that separates material appearance into components related and unrelated to weathering effects. Such a decomposition enables high-level appearance editing. We experiment with several linear and non-linear approaches, and achieve successful results from preliminary tests. Third, we introduce a method for capturing and transferring weathered textures to novel objects. As an initial proof-of-concept study, we capture the appearance of real surfaces drying, reduce the drying history to two parameters, then relate these spatially-varying control parameters to the object geometry and the shape of the original wetted region. We use these relationships to generate time-varying, spatial patterns of drying on synthetic objects. Finally, we present a comprehensive method for capturing progressively variant textures and the relevant context parameters. By relating textures and context parameters we are able to transfer the textures automatically to novel objects. We describe examples of capturing chemical effects, mechanical effects, and biological effects. We also demonstrate a user interface that provides a method for specifying where an object is exposed to external agents. We show the results of complex, geometry-dependent textures evolving on synthetic objects.