Image-Based Rendering is a 3D scene representation technique, whereby virtual viewpoints of a scene are rendered from a multi-view image set. The goal of IBR is to generate a feel of 3D from a limited number of input images, whilst maintaining a good rendering quality.
The central concept behind IBR is that each image consists of a set of light rays travelling from the scene to the camera. The light rays in question are modelled using the plenoptic function. Using this model, each image with finite resolution becomes a sample of the plenoptic function. Thus the rendering of a new viewpoint, in IBR, becomes the reconstruction of the plenoptic function from its samples.
We are interested in using multi-dimensional sampling theory to reduce the overall image set (i.e. number of spatial samples). In particular, we examine the bandwidth of the plenoptic function for a scene approximated with planar facets and bandlimited texture.