Abstract
With image-based relighting (IBL), one can render realistic relit images of a scene without prior knowledge of illumination environment. In this thesis, we present a framework for rendering from arbitrary viewpoints and relighting under arbitrarily rotated illumination environment of a real object from a set of images and a geometric model of the object under static illumination condition. Instead of using analytical reflectance model or using fixed viewpoint image-based relighting techniques, a clustering method for irregular and incomplete data is proposed to remove statistical redundancies and to extract illumination information from a large set of partial data. Further, we integrate the precomputed radiance transfer (PRT) technique into our framework to relight the scene that dynamically changing and constrained lighting condition. By doing so, we relax the limitation of rendering under the static illumination and make it possible to relight virtual objects realistically in real time.