Many applications of volume visualization require the display of
object boundaries. Using the presented compact volume representation, volume
visualization becomes feasible even over the Internet, while still
providing full spatial accuracy of the data. Representing just the boundary voxels
of objects dramatically reduces the amount of data to be transmitted
or stored. By exploiting known properties of the boundary voxels
(like spatial coherence and inter-voxel connectivity) the data
is further compressed. The resulting data representation is smaller by
a factor of 20-250 than the volume compressed with gzip. The
location of voxels within the volume is compressed very efficiently to about 2
bit/voxel. The compression rates for gradient data are lower, in the range of
3-8 bit/voxel, as gradient data is derivative information compared to
the original data, and thus containing less spatial coherence. Using a proper
gradient reconstruction scheme, gradients can be
estimated from voxel positions only, allowing to display objects just
after the well-compressed position data has arrived, instead of
waiting for the original gradient information.