Long before computers were invented artists and engineers, especially architects, were able to manually generate views of a scene having available not more than a few reference images. The automation of this intuitive ability is the target of today's image based modelling techniques.
In the image based approach the world is modelled in terms of a number of source images which are used to synthesise novel images that represent the scene from arbitrary viewpoints. As images can capture even the fine details of appearance, the complexity of the modelling process is decoupled from the complexity of the scene. Thus the amount of photorealism in the generated images only depends on the quality of the input views.
Image based modelling has evolved from research in both computer vision and graphics fields. Although the majority of the work in this area has focused on direct view interpolation between example images another category of methods approximate the plenoptic function [1] and directly synthesise images from arbitrary viewpoints. These two main categories are further discussed in the following section.