Image Processing Reference
In-Depth Information
39
Where Shall We Go Next?
39.1
Back Home Again
If you have read this far, you will have realized that there is a lot of fine detail to under-
stand when compressing video. You need to know about many things other than the com-
pression process because the bulk of your work happens before you compress a single
frame.
If you take each individual piece of knowledge separately, none of the steps involved
is very complex. It's just that there are rather a lot of them and if you compress your video
and leave out some of the preprocessing tricks or do them in the wrong order, it degrades
the quality of your output.
This concluding chapter looks beyond compression to examine some topics that crop
up in advanced work. It also ties up a few loose ends and looks at some speculative ideas
and new technologies that are not yet implemented but could be interesting avenues of
research.
The emerging high-performance computing systems built with massively parallel
clusters offer huge performance benefits and this leads to new ways to approach the com-
pression process with distributed algorithms. An example of a massively parallel cluster
is the one built by Virginia Tech. That system is a cluster of 1100 dual-processor Macintosh
G5 computers connected together with a very fast switched network and running some
software in addition to the underlying operating system so that work can be shared across
as many nodes as necessary. The Virginia Tech system was, for a short time, the 3 rd fastest
computer on the planet.
Software developers are already considering how the video-compression task should
be divided and distributed in such a clustered system. Taken to its extreme, we might
imagine that a 1000-node cluster would compress video 1000 times as fast as a single node.
There is the problem of splitting, distributing, and then gathering the results together, but
even if this slows the process somewhat, the gains are still likely to be profound. Breaking
the video into sequences based on the number of frames in a closed GOP might be a way
forward. It simplifies the distribution process as well as the re-aggregation of the coded
output.
619
Search WWH ::




Custom Search