Image Processing Reference
The MPEG-4 standard sees the video as a texture that is painted onto the surface of an
object. Other than that, the video is coded in a similar way to the MPEG-2 coding process
using DCT algorithms.
Simple shapes such as a rectangle are straightforward and already supported. It is only
necessary to indicate the size of the rectangle required. There is a minor limitation that the
size of the textured rectangle must be an integer multiple of the size of a macroblock.
The shape of the object that the video is being painted onto may change the visual
appearance. This is not the same as alpha shape-coded video. Rather, it is analogous to
projecting a film image onto the side of a building. Any overlap will spill over the sides
but it is nevertheless still being projected.
Sprites are used as static background objects, and they may be much larger than the view-
ing rectangle. The sprite image is delivered early on in the stream and then it can be
panned and scanned as necessary to use it as a background image. Some animation sys-
tems treat sprites as cell-animated character overlays but this is not how sprites appear to
be used in MPEG-4. That kind of animation is dealt with separately.
Certain efficiency techniques are available to deliver only those parts of the sprite
that are needed, when they are needed. Several of the tools provide a bandwidth-analysis
feature to help with this. Envivio 4Mation (originally known as Broadcast Studio Author)
has this built in. LiveStage Pro has a tool called Deliverator to do the same sort of thing
although that only operates on QuickTime wired movies at present.
Video content can be texture mapped onto a moving mesh. Several common examples
exist where a moving fish is modeled as a mesh. This mesh is spatially distorted while
playing back a video picture on the tiled surface.
Face and Body Animation
Face and body animation (FBA) is designed to support avatars and other similar models
and is a special case of the mesh-based texture-mapping technique. Using the facial ani-
mation specified by MPEG-4 would allow video conferencing to take place at bit rates of
1 Kbps or better.
Version 2 of MPEG-4 provides the body-animation support, and significant data
reduction is possible by using these structured techniques to recreate the desired scene.
This is well outside the territory of simple compressed video and more in the domain
of computer-animated special effects. Display systems don't often have the necessary
computational capability to support photorealistic modeling of this structured animation.
But that will come in due course.