Title of Thesis: Rendering of Multi-resolution Graphics Models Captured from Images Rendering of Multi-resolution Graphics Models Captured from Images

In both computer graphics and computer vision, scenes and objects must be represented mathematically in some way. Representations tailored to ease scene acquisition in vision, may be inefficient to render. Likewise, models designed specifically for efficient rendering may be difficult, if not impossible, to capture from cameras. In this thesis, we explore potential graphical models which are both convenient to acquire from images, and well suited for real-time rendering using modern graphics hardware. Our representation is a hierarchical hybrid image/geometry based model for graphics objects and scenes. Each of three scales of detail macro, meso, micro is represented differently in an attempt to most efficiently store, render and capture the model. Macro scale is the large scale or overall shape, which we represent using a low-resolution polygonal mesh. Meso scale detail, which is defined loosely as lying between macro and micro, is represented using displacement mapping, for which a novel hardware accelerated rendering algorithm is presented. At the micro scale, light-surface interaction properties are represented using a parameterized texture mapping technique called dynamic texture, where a linear texture basis is precomputed from example images and blended when rendering to reproduce different lighting conditions. A system was built and used to capture this type of model from real-world objects, which were then rendered, and integrated into virtual scenes. Additionally, experiments were performed which evaluate each subsystem independently.