ABSTRACT
The purpose of image-and video-based rendering (IVBR) is to be able to synthesize photo-realistic new, virtual views of real-world scenes and events from no more than a set of conventional photographs or videos of the scene. Purely image-based, or plenoptic, approaches densely sample scene appearance using a large number of images (Sections 17.2 and 5.3). New views are generated by simply re-sampling the captured image data. In contrast, geometry-assisted methods require much less input images. Here, 3D scene geometry is either a priori known, reconstructed from the captured imagery, or acquired separately by some other means, e.g., ranging imaging (Chapter 4). With (approximate) scene geometry available, views from arbitrary viewpoints are synthesized using the acquired images as geometry texture (Section 17.3). Over the years, various IVBR methods have been proposed that can all be categorized in-between these two limiting cases [Lengyel 98] (Figure 17.1). Their respective advantages and limitations have been discussed in several IVBR surveys [Shum and Kang 00, Smolic et al. 09, Linz 11, Germann 12].
