Virtual Environments and Computer Graphics (VECG)
Department of Computer Science
UCL

 

The Virtual Light Field

EPSRC Project GR/R13685/01

Researchers:

Jesper Mortensen, PhD Student
Pankaj Khanna, Research Fellow (p/t PhD Student)
Insu Yu, Research Fellow, (p/t PhD student)

Principal Investigator:

Mel Slater

An ideal light field is a function L(p,w,t) that gives the radiance at every point p for every direction w at time t in a space, also called a lumigraph. Complete knowledge of this function for a scene would provide a total solution to the problem of computer graphics rendering of the scene, as it changes through time. 

In practice the function L is sampled with a discrete set of straight line rays at a specific time instant. This sample of rays, each carrying a radiance value, is what is usually referred to as a 'light field' [1][2]. In practice such a discrete light field is typically obtained from a large set of images, created from known camera positions. The technique is used most commonly with digitally captured images of real scenes, and the light field can then be used to render these scenes from camera orientations and positions not included in the original set. The technique may also be used for synthetic scenes - where a large number of images are created using a rendering system such as ray tracing, and then new interpolated images can be reconstructed from the light field. When used for synthetic scenes we refer to a virtual light field. This term was probably first used by Gortler et. al. in [1].

Here the term 'virtual light field' is extended to mean a rendering system where the light field is constructed for a synthetic scene, but without the use of a separate rendering system. The scene is 'rendered' into a light field in such a way that the radiance values are computed directly during the construction of the light field itself. It turns out that this approach supports the creation of a light field function which intrinsically generates  global illumination, with a mix of specularly and diffusely reflecting materials. Moreover, having created such a light field it can be rendered into a 2D image either in the traditional way through a pin-hole camera, or by allowing the rays to pass through a lens system forming an image on an image plane - thus including depth of field effects.

A first attempt to implement a virtual light field approach is described in [3]. That approach resulted in poor images, which nevertheless showed the promise of the technique for global illumination. The major problem was the massive memory required, and the lack of accuracy in visibility of rays between objects. 

Since then several iterations of the algorithm and implementation have been carried out, with a vast increase in quality and performance. 

This research is supported by an  EPSRC Senior Fellowship Award, and an EPSRC project.

[1] Gortler, S., Grzeszczuk, R., Szeliski, R., Cohen, M. (1996) The Lumigraph, Computer Graphics (SIGGRAPH), Annual Conference Series, 43-52.

[2] Levoy M and Hanrahan, P. (1996) Light Field Rendering, Computer Graphics (SIGGRAPH), Annual Conference Series, 31-42.

[3] M. Slater (2000) A Note on Virtual Light Fields (PDF), Department of Computer Science Research Note RN/00/26/ April 5th 2000, University College London.

Further Resources

Some papers and video material