Lucas Pereira
						David Redkey
						cs348c Project Proposal
						February 13, 1996

		
		Depth Information for Light Fields


	For our project we propose that we will investigate methods
for the generation and use of depth information in light fields.  We
will attempt to generate the depth information for light fields acquired
from a static real-world scene.  We will use this depth information to
combine the light field with synthetically rendered polygon models,
thus merging the real and virtual worlds in a new way.  Our overall
goal is to assess the feasibility, benefits, and limitations of this
approach.


PART I: Depth Generation
------------------------

	Our goal is to be able to generate depth values for any
arbitrary light field, whether it has been collected from the real
world or synthetically generated.  To help us test the accuracy of our
depth reconstruction, we can record the Z values for a synthetically
generated light field.  Then we can compare our generated values to
the actual Z values, and use this to gauge the effectiveness of our
algorithms.

	The depth-information generation step will involve expanding
on the work of current computer-vision research, such as Prof. Carlo
Tomasi's vision system.  However, with light fields we have several
key factors that should make our algorithm simpler and more robust.
First, we know the exact position and orientation of the camera, so we
do not have to do pose estimation.  Secondly, we have access to the
view from virtually any camera direction within the light field.
Thus, rather than trying to reconstruct depth from only two images, we
can find the depth of each ray from a whole contiguous plane of
images.  The coherence of these images should make it easier for us to
calculate depth reliably, and be able to recognize and throw away
artifacts due to occlusion from other viewpoints.

	If we find that we are unable to reproduce accurate Z values,
we may just use the depth values from a synthetically generated
light field (i.e. store the Z-buffer values for each rendering).  This
will allow us to test PART II.


PART II: Using Depth Information
--------------------------------

	We also wish to explore the applicability of depth information
to combining light fields with other representations, such as
polygonal objects.  We will examine the feasibility of this approach,
and find out if there are any major problems with depth aliasing or
other artifacts.

	For our final presentation, we will generate a (possibly 
interactive) display that shows how the light field can be merged
with other objects.  This display should demonstrate how well our
approach simulates proper occlusion between the real (light-field)
and virtual (polygon) world.

	Time permitting, we would also like to make a stereo display
(either on the workbench or a stereo monitor), to give the viewer a
better feel for how well the real and virtual worlds merge in 3D.