Gary,

what you are after happend to be the most dramatic change in MPEG-4 
compared to previous standards. Those considered a movie to be a 
sequence of rectangular images, all the same size. In contrast, MPEG-4 
thinks of a movie as a composition of objects (using a VRML inspired 
scene discription "language"), each of which is a sequence of irregular 
images.

Obvious decoding such a system is easy - expand each object and layer 
them one over the other. Encoding is easy for synthetic material, like 
a computer generated cartoon. Encoding existing video is *very* hard 
since it would involve segmentation, recovering hidden pixels and other 
problems that happen to be interesting to you. When the standard was 
created in 1999 it was hoped that these problems would be solved 
eventually and the decoders would be ready. I don't know what the 
popular MPEG-4 encoders, such as DivX;), do but I imagine they don't 
try to separate the images at all.

In another email you asked about separating MP3s into tracks. With 
recordings that put some instruments on one channel, the rest on the 
other one and the vocals in the middle (equally on both) you might be 
able to separate the voice with a little signal processing.

Good luck,
-- Jecel