Page 91 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

68 Chapter 3. Video Coding: Standards

which moving parts of a scene can be represented while maintaining a higher-
resolution representation in more static areas of the scene.
The syntax of the bitstream in this mode is identical to the syntax for cod-
ing without the mode, but the semantics, or interpretation of the bitstream,
is somewhat di1erent. In this mode, the portion of the picture covered by a
macroblock is twice as wide and twice as high. Thus, there is approximately
one-quarter the number of macroblocks as there would be without this mode.
Motion vector data also refers to blocks of twice the normal height and width,
or 32 × 32 and 16 ×16 instead of the normal 16 ×16 and 8 ×8. For example,
the decoder receives and decodes a 16 ×16 DFD block at the reduced resolu-
tion. The decoder then upsamples this block to 32 × 32 at the higher resolution.
The decoder then upsamples the received motion vector by a factor of 2 and
uses it to produce a 32 × 32 prediction from the reference picture. The DFD
block and the prediction block are then added to produce a 32 × 32 block at
the higher resolution.

3.4.6.12 Independent Segment Decoding Mode (Annex R)
This mode allows a picture to be constructed without any data dependencies
that cross video picture segments. Thus, this mode provides error robustness
by preventing the propagation of erroneous data across the boundaries of video
picture segments.
In this mode, a video picture segment can be a slice, a GOB or multi-GOBs
with nonempty GOB headers, or a complete picture. When this mode is in use,
the video picture segment boundaries are treated as picture boundaries. In other
words, each video picture segment is decoded with complete independence
from all other video picture segments, and is also independent of all data
outside the corresponding video picture segment in the reference picture(s).
For example, motion vectors of blocks outside the current video picture
segment cannot be used when calculating the current motion vector predictor.
Similarly, motion vectors of blocks outside the current video picture segment
cannot be used as remote motion vectors for overlapped block-motion compen-
sation when the Advanced Prediction mode is in use. In addition, no motion
vectors are allowed to reference areas outside the corresponding video picture
segment in the reference picture(s).

3.4.6.13 Alternative INTER VLC Mode (Annex S)
This mode improves the e,ciency of encoding some INTER macroblocks
by allowing a VLC table originally designed for INTRA macroblocks to be
used for some INTER macroblocks. The INTRA VLC table used in the ad-
vanced INTRA coding mode (annex I) is designed to e,ciently encode INTRA
blocks. Thus, it is optimized for coding blocks with many large-valued coe,-