Page 107 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

84 Chapter 3. Video Coding: Standards

or categories. The VLC codeword indicates to which category the encoded
di1erence belongs, whereas the FLC codeword, then, uniquely identi/es the
di1erence within that category. Instead of this special treatment, the INTRA
DC coe,cients can optionally be encoded using the same INTRA AC VLC
table described next. To achieve compatibility with H.263, the INTRA DC
coe,cients can also optionally be encoded without prediction using an 8-bit
FLC codeword.
All other coe,cients are encoded using a procedure similar to that of H.263.
Thus, the scanned quantized coe,cients are converted into an intermediate
set of EVENTS of the form (LAST, RUN, LEVEL). The most commonly
occurring events are then encoded using standard VLC tables. There are two
standard VLC tables: one for INTRA blocks and another for INTER blocks.
To achieve compatibility with H.263, the VLC table for INTER blocks can
optionally be used for both INTER and INTRA blocks. Less frequent EVENTS
are encoded with the help of an ESCAPE codeword.

3.5.5 Still-Texture Coding

The MPEG-4 also supports coding of static textures (or still images). This
mode uses subband coding based on the discrete wavelet transform (DWT).
As discussed in Section 2.6.3, the DWT is used in subband coding to
apply a nonuniform decomposition (refer to Figure 2.12(b)) to the texture
information. This results in a decomposition tree of subbands. The lowest
subband (horizontal low=vertical low (LL)) is known as the DC subband,
whereas the remaining subbands are known as the AC subbands.
In MPEG-4, the DWT can be either a Joating-point or an integer transform,
as signaled by the encoder in the bitstream. The encoder can also choose to
use a set of default /lters or to use its own /lters and de/ne them in the
bitstream.
The quantized coe,cients of the DC subband are encoded using DPCM
followed by arithmetic coding. The choice of the predictor for a particular
coe,cient depends on the magnitude of the horizontal and vertical gradients of
neighboring coe,cients. If the horizontal gradient is smaller than the vertical
gradient, then prediction is performed using the left neighboring coe,cient;
otherwise the top neighboring coe,cient is employed.
The quantized coe,cients of the AC subbands are encoded using a zero-tree
algorithm followed by arithmetic coding.

3.5.6 Sprite Coding
An interesting mode supported by MPEG-4 is sprite coding. A sprite consists
of those parts of an object that are present in the scene throughout a video