Public Software Release
                           -----------------------

                    FULLY SCALABLE, LOW-LATENCY VIDEO CODEC
              BASED ON 2 & 3 DIMENSIONAL SUBBAND TRANSFORMATION
          AND INTER-FRAME PROGRESSIVE CODING OF SUBBAND COEFFICIENTS
WITH SCALING FOR CONSTANT BIT RATE (CBR) OR CONSTANT DISTORTION (VBR)
CRITERION

                                     by

                              David S. Taubman
                    University of California, Berkeley

                           -----------------------

I am pleased to announce the release of scalable video compression and
decompression software.  This software may be regarded as a sequel to
our previously released scalable image and video codec (June 30,
1994).  Both the new codec and the previously released codec are available from
        http://www-video.eecs.berkeley.edu/download/scalable

Unlike conventional compression standards, these codecs perform fully
scalable compression of both video and images.  This means that
subsets may be extracted from the compressed data stream so as to
satisfy virtually any constraint on bit rate and a wide range of
compatible display frame resolutions and frame rates.  That is, the
video is compressed once only, after which scaling may occur to meet
unpredictable demands imposed by storage, distribution and display
systems.

The new codec differs from our previously released software codec in
two primary respects:

1) The new codec exploits temporal redundancy by using both temporal
subband decomposition and inter-frame progressive coding techniques,
rather than temporal subband decomposition alone.  As a consequence,
fewer levels of temporal subband decomposition are required to achieve
good compression with the new codec than with the previously released
one.  As such, the new compression/decompression algorithm is able to
achieve good compression efficiency with much less memory and much
lower end-to-end delay than that implemented by our previous software
release.

2) Whereas our previously released software compression algorithm
generated a bit stream, the new software generates a packet stream.
The distinction between these two cases is that scaling entities must
be intimately familiar with the detailed bit stream syntax if scaling
is to take place within the context of a raw bit stream.  On the other
hand, the packet stream imposes a higher level abstraction on top of
this syntax, so as to enable simple, generic scaling entities to
select from a pre-defined set of bit rate targets.  Simple headers in
the packet stream provide sufficient information to perform scaling
with either a constant bit rate or constant distortion criteria.
These concepts are explained in the technical paper included with the
software release.

Note: All correspondence should be directed to:
 avz@eecs.berkeley.edu

For suggestions or questions regarding envisaged hardware or
software implementations, algorithm enhancements or further
developments, feel free to contact Professor Avideh Zakhor at
 avz@eecs.berkeley.edu

The `scalable2.tar.Z' file contains: software and documentation; a
technical paper, discussing the theoretical aspects of the
algorithm, scaling operations and experimental performance results;
and examples.  Software is driven by a general configuration language,
enabling numerous compression algorithms to be generated within the
scalable context.  Decompression with compatible, but not identical
configurations enables frame resolution and frame rate scalability.
Although frame rate and frame resolution scalability are primarily of
value for compatible decompression, bit rate scalability is of value
throughout the path from compression, possibly to storage, through
distribution and eventually to decompression.  Bit rate scalability is
currently only implemented within the decompression software, so that
bit rate constraints may be imposed immediately before decompression.
We note, however, that the packetization scheme enables simple, generic
rate scaling operations to be implemented anywhere in the distribution
path taken by the scalable data stream.  The decompression software
may be used in conjunction with a separate viewing console with
VCR-like controls for video playback, rewind, fast forward, etc., on
high end workstations.  The viewing console software itself is not
included in `scalable2.tar.Z', but may be recovered from the
`scalable.tar.Z' file containing our previously released scalable codec.

Initial investigations suggest the following:
 For video compression of progressively scanned video
        sequences, the compression efficiency is comparable to that of
        the non-scalable MPEG-1 standard.  Generally, performance is
        significantly better than MPEG when camera motion conforms to
 a pan model (i.e. still, jitter, and true pan), regardless of
        the amount of foreground motion.  On the other hand,
        performance can be significantly worse (2-3 dB) than MPEG when
        camera motion is zoom or translation and when there is little
        or no complex foreground motion.

It is important to appreciate, however, that the primary feature of
this suite of compression algorithms is scalability.

Computational complexity appears to be perhaps several times that of
MPEG, however the algorithm is inherently amenable to highly parallel
hardware or software implementation.

Feel free to experiment with the software and send us feedback
regarding your experiences and/or application ideas.

        David Taubman                           (August 29, 1994)