Parallel Video Compression
In MPEG video compression, the motion compensated
prediction step is very computationally intense.
Motion compensation is important because it can provide a lower
data rate than using just intraframe approaches for a fixed image
quality. In this project we explore the use of high performance
parallel computers to compress video sequences.
Two types of parallelism can be used for MPEG: spatial and temporal.
In this work we have investigated both approaches.
We found that the large amount of data exchanged between
the processors and the disk system caused
an I/O bottleneck. An I/O queueing algorithm was developed to
reduce the I/O contention. Faster than real-time MPEG encoding has
been achieved.
The following figure is a comparisson of the
performance of the algorithm with and without the I/O queue.
Detail
information may be found in:
Notice: the copyrights to the following papers are held by the
publishers. The attached PostScript and PDF files are preprints. Please treat
this material in a way consistent with the "fair use'' provisions of
the appropriate copyright laws.
Many of these papers include source code and images used in the work.
The README files in each directory at the
ftp site
provide the needed details
including the correct citations for each paper.
Address all comments and questions to
Professor
Edward J. Delp.
K.Shen and E.J. Delp,
"A parallel implementation of an MPEG encoder: Faster than real-time!,"
Proceedings of the SPIE Conference on Digital Video Compression: Algorithms
and Technologies, February 5-10, 1995, San Jose, California,
pp. 407-418.
The
readme file ,
compressed postscript file,
PDF file,
and the
ftp site.
Source code is available
for the Intel
Touchstone Delta and Intel
Paragon.
K. Shen and E. J. Delp,
"A Spatial-Temporal Parallel Approach For Real-Time
MPEG Video Compression,"
Proceedings of the
25th International Conference on Parallel Processing,
August 13-15, 1996,
Bloomingdale, IL, pp. II100-II107.
The
readme file,
compressed postscript file,
PDF file, and the
ftp site.
Professor Edward J. Delp