Page 84 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 3.4. The H.263 Standard 61

using the horizontal DC and AC option, a new alternate vertical scanning
pattern (Figure 3.5(b)) is utilized.
4. The quantized INTRA coe,cients are encoded using a new VLC table
optimized for the global statistics of INTRA macroblocks.

3.4.6.4 Deblocking Filter Mode (Annex J)
In this optional mode, a /lter is applied, both at the encoder and at the de-
coder, across the boundaries of luma and chroma 8 × 8 blocks of reconstructed
pictures before storing them in the picture memory. In other words, the /lter
a1ects the picture that is used for the prediction of subsequent pictures and
thus lies within the motion prediction loop.
The deblocking /lter operates using a set of four pel values either on
a horizontal or on a vertical line of the reconstructed picture. Two of the
four pels belong to one block, whereas the other two belong to a neighboring
block. The weights of the /lter’s coe,cients depend on the quantizer step size,
where stronger coe,cients are used for a coarser quantizer, and vice versa.
No /ltering is performed across a picture edge. Similarly, when the Inde-
pendent Segment Decoding (ISD) mode is in use, no /ltering is performed
across slice edges (when the Slice Structured mode is in use) or across the top
boundary of GOBs having GOB headers present (when the Slice Structured
mode is not in use). When this mode is used together with the Improved PB-
frames mode, the backward prediction of the B-macroblock is based on the
reconstructed P-macroblock before the deblocking edge /lter operations. The
mode applies only for the P-, I-, EP-, or EI-pictures or the P-picture part of
an Improved PB-frame. Possible /ltering of B-pictures or the B-picture part
of an Improved PB-frame is not a matter for standardization.
In addition to the /ltering operation, this mode allows the use of four
motion vectors per macroblock and also the use of unrestricted motion vectors.
This mode improves the prediction quality and signi/cantly reduces blocking
artefacts.

3.4.6.5 Slice Structured Mode (Annex K)
In this optional mode, a slice layer is employed instead of the normal GOB
layer. This mode is used to provide enhanced error resilience, to make the
bitstream more amenable to use with packet-based networks, and to minimize
video delay. A slice layer allows a Jexible partitioning of the picture into
segments containing a variable number of macroblocks. It also allows more
control over the shape of segments. In addition, a slice structure provides more
Jexibility in the transmission order. This is in contrast with a GOB layer,
which only allows partitioning into /xed-size, /xed-shape segments with /xed
transmission order.