November 30, 1995
Abstract -
Digital video coding is currently a major topic
of research since high levels of compression are necessary
to develop a large number of important video products
such as video telephone, video conferencing, video databases,
and high definition television (HDTV).
In advanced video coders (e.g. MPEG), motion vectors are
used to improve the prediction of the frame to be coded.
Then, these motion vectors and the prediction error or
difference frame must be encoded with bits. It is clear
that increasing the number of motion bits decreases the
number of bits needed to encode the difference frame,
but more analysis is needed to find the optimal motion
bit allocations that minimize the total bit rate for real
video sequences. In fact, the number of bits allocated
to the motion vectors in most video coders is based on
heuristics and empirical experiments, and typically is just large
enough to encode the motion vectors with 1 or 1/2 pixel
accuracy, and all motion vectors are encoded with the
same accuracy. In this work, we explore the benefits
of encoding the motion vectors with other accuracies,
and the benefits of encoding different motion vectors
with different accuracies in the same frame. To do this,
we analytically model the effect of motion vector accuracy
and thereby obtain expressions for the total bit rate in
terms of the accuracies, for typical lossless
and lossy video coders. Minimizing these expressions leads
to simple procedures for determining how accurately to
encode the motion vectors. We implement the video
coders, present experimental results of their performance
on real video sequences, and discuss the benefits of our
procedures. Finally, we discuss the generality of
our work and its potential applications to other
motion-compensated video coders.