Databases Reference
In-Depth Information
T A B L E 19 . 8
The standardized H.263 formats.
Picture format
Number of lu-
minance pixels
(columns)
Number of
lu-
Number
of
Number
of
minance
lines
chrominance
pixels (columns)
chrominance
lines (rows)
(rows)
sub-QCIF
128
96
64
48
QCIF
176
144
88
72
CIF
352
288
176
144
4CIF
704
576
352
288
16CIF
1408
1152
704
576
by the standard are shown in Table
19.8
. The picture is divided into
groups of blocks
(GOBs)
or slices. A group of blocks is a strip of pixels across the picture with a height that is a multiple
of 16 lines. The number of multiples depends on the size of the picture, and the bottommost
GOB may have less than 16 lines. Each GOB is divided into macroblocks, which are defined
as in the H.261 recommendation.
A block diagram of the baseline video coder is shown in Figure
19.16
. It is very similar
to Figure
19.10
, the block diagram for the H.261 encoder. The only major difference is the
ability to work with both predicted or
P
frames and intra or
I
frames. As in the case of
H.261, the motion-compensated prediction is performed on a macroblock basis. The vertical
and horizontal components of the motion vector are restricted to the range
.The
transform used for representing the prediction errors in the case of the
P
frame and the pixels
in the case of the
I
frames is the discrete cosine transform. The transform coefficients are
quantized using uniformmidtread quantizers. The DC coefficient of the intra block is quantized
using a uniform quantizer with a step size of 8. There are 31 quantizers available for the
quantization of all other coefficients with step sizes ranging from 2 to 62. Apart from the DC
coefficient of the intra block, all coefficients in a macroblock are quantized using the same
quantizer.
The motion vectors are differentially encoded. The prediction is the median of the motion
vectors in the neighboring blocks. The H.263 recommendation allows half pixel motion
compensation as opposed to only integer pixel compensation (which is all that is allowed in
H.261). Notice that the sign of the component is encoded in the last bit of the variable length
code, a “0” for positive values and a “1” for negative values. Two values that differ only in
their sign differ only in the least significant bit.
The code for the quantized transform coefficients is indexed by three indicators. The first
indicates whether the coefficient being encoded is the last nonzero coefficient in the zigzag
scan. The second indicator is the number of zero coefficients preceding the coefficient being
encoded, and the last indicates the absolute value of the quantized coefficient level. The sign
bit is appended as the last bit of the variable-length code.
Here we describe some of the optional modes of the H.263 recommendation. The first four
options were part of the initial H.263 specification. The remaining options were added later
and the resulting standard is sometimes referred to as the H.263
[−
16
,
15
.
5
]
+
standard.
