Media Compression - Scalable Continuous Media Streaming Systems - page 33

media stream will present additional complexities in terms of scheduling, I/O, and resource
allocation. We will return to this issue in Chapter 6 when we introduce bit-rate smoothing.
2.5 Media Adaptation
Once a media stream is compressed, it is typically stored in a media server awaiting user
requests for streaming. The media server typically will not further process the media data prior
to streaming. In some cases, however, the ability to reshape the media stream can be very
desirable.
Take media streaming in the current Internet as an example. As the Internet is a best-effort
network, it cannot provide any guarantee on bandwidth availability. Thus if the available
network bandwidth drops below the encoded media bit-rate, then the media streaming session
will often be disrupted - not uncommon in today's Internet.
In another scenario, a public media server may need to serve the same content to clients
with different bandwidth availability, e.g., some may be connected via ADSL (1.5 Mbps) while
others may have high-speed connections. The service provider commonly will either encode
the media according to the lowest bit-rate of their target users (thus sacrificing quality for
users with better connections) or encode multiple versions of the same content at different
bit-rates for the users to choose. This latter approach is costly as the encoding process is often
labor-intensive and takes considerable time.
The previous two scenarios in fact are due to the same fundamental problem - matching the
media bit-rate to the bandwidth available. One emerging solution is to use layered video coding
where a video is encoded into one base layer and a number of enhancement layers [39-42]. The
base layer provides the lowest bit-rate video (with the lowest visual quality) while adding each
enhancement layer will progressively improve the visual quality. With layered video coding,
the server can then adjust the number of layers to transmit according to the bandwidth available
or according to the user's connection speed.
Another solution is video transcoding. A video transcoder can convert a compressed video
from a high bit-rate stream to a low bit-rate stream by selectively dropping information in
the process. Unlike layered video coding, the output bit-rate is continuously adjustable, and it
does not require any modification to the decoder for playback. Common transcoding techniques
are requantization [43-44], spatial downscaling [45-48], and temporal downscaling [49]. Each
technique has different ranges of achievable bit-rate reductions (see Section 2.5.2) and hence the
choice of transcoding techniques will depend on the amount of bit-rate reduction required [50].
In the following, we investigate the achievable rate reductions using requantization and
spatial downscaling based on the MPEG1 compression standard. Experiments reveal that
neither requantization nor spatial downscaling alone can achieve a sufficiently wide range
of bit-rate reductions. Instead, by combining them and selecting the appropriate transcoding
techniques we can achieve a wide range of output transcoded bit-rate from 100% down to 20%
of the original media bit-rate.
2.5.1 Transcoding Techniques
In this session, we review requantization and spatial downscaling, two existing algorithms for
performing video transcoding.