Digital Signal Processing Reference
In-Depth Information
15.2 H.264 Video Compression Standard
This trend of more profiles and levels continues in H.264. The
complexity and processing requirements increase dramatically as
H.264 can compress approximately twice as much video as
MPEG-2, at similar quality levels. Alternatively, H.264 can deliver
substantially better quality at the same bit rate as MPEG-2.
Many of the key H.264 enhancements are listed below, fol-
lowed by a more detailed discussion of some of these enhance-
ments. For those interested in deeper discussion of the
implementation details and options on H.264, there are a number
of H.264 texts written at a level for video engineers, and of course,
there is the H.264 standard itself.
15.2.1 H.264 Levels and Pro
les
The levels and profiles have been expanded to enable new
options. Table 15.4 summarizes these. The levels provide for
maximum permissible constraints for various parameters, which
drive the processing power, memory and other requirements of
the encoder and especially decoder.
The profiles include baseline constrained, baseline, extended
baseline, main and various high-quality broadcast-level profiles.
Some of the features defined in the profile description are
explained in the following sections.
15.2.1.1 H.264 Profiles
Constrained Baseline Pro
le (CBP)
Mainly found in low-cost applications, this profile is most
typically used in videoconferencing and mobile applications. It
has the minimum feature set. It was released after the Baseline
profile, and is intended to further eliminate features not required
in specific applications, such as redundant slices, arbitrary slice-
ordering and flexible macroblock ordering.
Baseline Pro
le (BP)
This profile is used primarily in low-cost applications where
there may be data loss during transmission, such as mobile
devices or video conferencing. It is similar to CBP, but has
features to help compensate for data loss or corrupted frames.
Baseline profile is also intended for low latency applications. It
includes I- and P-frames, but excludes B-frames, which require
more latency to support the frame buffering required by back-
wards prediction. Baseline profiles use the 4
4 DCT transform
