Page 82 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 3.4. The H.263 Standard 59

in the picture header, then the following modi/cations apply:
1. The motion vector range no longer depends on the motion vector
prediction value. There are two cases here:
(a) If the UUI data /eld in the picture header is set to “1,” the mo-
tion vector range depends on the picture format. For standard-
ized picture formats up to CIF the range is [−32; 31:5], for those
up to 4CIF the range is [−64; 63:5], for those up to 16CIF the
range is [−128; 127:5], and for even larger custom picture formats
the range is [−256; 255:5]. In addition, the horizontal and vertical
motion vector ranges may be di1erent for custom picture formats.
(b) If, however, the UUI data /eld is set to “01,” the motion vectors Q1
are not limited except by their distance to the coded area border,
as explained by the following restriction rule: the motion vector
values are restricted such that no element of the 16 × 16 (or 8 × 8)
region that is selected shall have a horizontal or vertical distance
more than 15 pels outside the coded picture area.
2. A new VLC table is employed to encode the motion vector di1erences.
This table has the following properties:

(a) The codes are single-valued. In other words, each codeword corre-
sponds to a single motion vector di1erence value. This is in contrast
to the double-valued VLC codes of version 1, where each code-
word can represent one of two possible motion vector di1erences.
Double-valued codes were not popular due to their high implemen-
tation cost and the limitations on their extendibility.
(b) The table employs reversible variable-length coding (RVLC) code-
words. Such codewords can be decoded in both the forward and
backward directions. As discussed in Chapter 9, the use of RVLCs
can increase the error resilience of video bitstreams. In addition,
RVLCs are easier to implement because they can easily be gener-
ated and decoded using a simple state machine.

3.4.6.3 Advanced INTRA Coding Mode (Annex I)
This optional mode signi/cantly improves the compression performance when
coding INTRA macroblocks. The mode is applied both to INTRA macroblocks
within INTRA-pictures and to INTRA macroblocks within INTER-pictures.
The improved compression performance of this mode is achieved as follows:

1. INTRA blocks are predicted from their neighboring INTRA blocks.
Block prediction always uses data from the same luma or chroma