Page 92 - Video Coding for Mobile Communications Efficiency, Complexity, and Resilience

Section 3.4. The H.263 Standard 69

cients and small runs of zeros. There are cases, however, where the statistics
of INTER blocks can approximate the statistics of INTRA blocks. This is par-
ticularly possible when signi/cant changes are evident in the picture or when
small quantizer step sizes are employed. In such cases, it can become more
e,cient to encode INTER blocks using the INTRA VLC table.
In this mode, the encoder would normally choose to use the INTRA VLC
table for coding an INTER block only when the use of this table results in
fewer bits than the use of the INTER VLC table. This use of the alterna-
tive INTRA VLC table, however, is subject to the condition that the decoder
would be able to detect which of the two tables was used for encoding. Thus,
the alternative INTRA VLC table can be used, subject to the condition that
decoding using the INTER VLC table would result in runs of zeros so long
as to indicate the presence of more than 64 coe,cients in the block.

3.4.6.14 Modi>ed Quantization Mode (Annex T)
In this mode, the quantizer operation is modi/ed. In particular, this mode
includes the following four key features:
1. In normal mode, the change of the quantization parameter at the
macroblock level is limited to a maximum of ±2. This mode, however,
improves the bit-rate control ability by allowing the quantization param-
eter to be changed at the macroblock level to any of its 31 permissible
values.
2. In normal mode, the chroma quantizer step size is the same as that for
luma. This mode, however, improves the /delity of chroma by specifying
a smaller quantizer step size for chroma than that for luma.
3. The true value of a DCT coe,cient prior to quantization can be as
high as 2040. Thus, when the quantization parameter is less than 8,
the quantized DCT coe,cients can be outside the range [−127; +127]
permissible in the normal mode. Such coe,cients are clipped to the
permissible range before being encoded. This mode, however, extends
the range of representable quantized DCT coe,cient values to allow
the representation of any possible true coe,cient value to within the
accuracy allowed by the quantizer step size.
4. In this mode certain restrictions are placed on the encoded DCT
coe,cient values to improve the detectability of errors and to minimize
decoding complexity.
Kossentini et al. [71, 72] provide an excellent overview of H.263+ and
evaluate the performance of the modes individually and in di1erent combina-
tions.