Extended Light Fields

Extended Light Fields

Light field rendering is a new image based rendering method, i.e. a new image is produced from other existing images. This is fast and independent from the complexity of the scene, so rendering of complex objects becomes possible at interactive rates using this precalculation.


These are images rendered from a human head with traditional methods (Volpack) and an extended light field (head.lif in examples.tar.gz) that stores 256 such images of size 256x256 at 8 parameter combinations:

Volpack Light Field
VolPack, first view lightfield, first view
VolPack, second view lightfield, second view

Fine details are lost due to blurring caused by the interpolations, but in general the image is still a close approximation. Differences in the geometry are caused by the fact that light field rendering uses perspective projection whereas the original Volpack rendering only supports parallel projection. The application to volume rendering is discussed in more detail in the thesis.

The major drawback of light fields so far was the restriction to static scenes. Especially for volume rendering changing additional parameters like position and steepness of a ramp that maps function from voxel density to its opacity (shown in the top of the images together with a histogram) is important. Extended light fields overcome this limitation by storing several generic light fields and interpolating between them. An irregular sampling and non-linear basis functions are used to keep the number of generic light fields small.

This way it was possible to cover the interesting parameter area for this data with only 8 generic light fields with sufficient accuracy of the interpolation and to change the parameters interactively while looking at the extended light field, producing the images on the right.

lightfield with steepness 100 and position 15
lightfield with 58/10
lightfield with 40/10
lightfield with 30/1