Image-Based Rendering

Image-based rendering can be viewed as the fusion of computer vision (creating models from images) and 3D computer graphics (rendering images from models), where we render images from images without explicitly creating a geometric model. To render from images, we require not only the photographic images of the scene, but range data on a pixel-by-pixel basis. Given the digital image and a corresponding range image, we can render the scene over a wide range, providing realistic renderings. Multiple images of the same scene from different locations allow renderings from virtually any position in the environment without ``holes'' created by occlusions. Acquiring digital photographs is commonplace with today's technology. Our photograph acquisition relies on digital cameras, such as the Canon EOS D2000 or Olympus DL300 camera, or scanned photographs. Acquisition of range data is not nearly as simple. The main point of this paper is to describe our active range acquisition system, contrasting it with other active and passive systems.