The majority of the work on image-based rendering has focussed on the problem of generating novel views from example images, with the restriction that the illumination is fixed. These approaches can be characterised by those that directly manipulate single or multiple images and those that reconstruct and use the plenoptic function.