This section describes the procedure for controlling the bit-rate of
the Test Model by adapting the macroblock quantization parameter
quantizer_scale. The algorithm works in three-steps:
1 Target bit allocation:
this step estimates the number of bits available to code the next
picture. It is performed before coding the picture.
2 Rate control: by means
of a "virtual buffer", this step sets the reference
value of the quantization parameter for each macroblock.
3 Adaptive quantization:
this step modulates the reference value of the quantization parameter
according to the spatial activity in the macroblock to derive
the value of the quantization parameter, mquant, that is used
to quantize the macroblock.
After a picture of a certain type (I, P, or B) is encoded, the respective "global complexity measure" (Xi, Xp, or Xb) is updated as:
where Si, Sp, Sb are the number of bits generated
by encoding this picture and Qi, Qp and Qb are the average quantization
parameter computed by averaging the actual quantization values
used during the encoding of the all the macroblocks, including
the skipped macroblocks.
Initial values
Xi = (160 * bit_rate) / 115
Xp = (60 * bit_rate) / 115
Xb = (42 * bit_rate) / 115
bit_rate is measured in
bits/s.
The target number of bits for the next picture in the Group of pictures (Ti, Tp, or Tb) is computed as:
Where:
Kp and Kb are "universal" constants dependent
on the quantization matrices. For the matrices specified in sections
7.1 and 7.2 Kp = 1.0 and Kb = 1.4.
R is the remaining number of bits assigned to the
GROUP OF PICTURES. R is updated as follows:
After encoding a picture , R = R - Si,p,b
Where is Si,p,b is the number of bits generated
in the picture just encoded (picture type is I, P or B).
Before encoding the first picture in a GROUP OF PICTURES
(an I-picture):
R = G + R
G = bit_rate * N / picture_rate
N is the number of pictures in the GROUP OF PICTURES.
At the start of the sequence R = 0.
Np and Nb are the number of P-pictures and B-pictures
remaining in the current GROUP OF PICTURES in the encoding order.
| R-bits | ||||||||||||
| Np = 3 | ||||||||||||
| Nb = 4 | ||||||||||||
Before encoding macroblock j (j >= 1), compute
the fullness of the appropriate virtual buffer:
or
or
…depending on the picture type.
where
are initial fullnesses of
virtual buffers - one for each picture type.
Bj is the number of bits generated by encoding all
macroblocks in the picture up to and including j.
MB_cnt is the number of macroblocks in the picture.
are the fullnesses of virtual
buffers at macroblock j- one for each picture type.
The final fullness of the virtual buffer (
:
j = MB_cnt) is used as 
for encoding the
next picture of the same type.
Next compute the reference quantization parameter Qj for macroblock j as follows:
where the "reaction parameter" r is given by
and dj is the fullness of the appropriate virtual
buffer.
The initial value for the virtual buffer fullness is:
Compute a spatial activity measure for the macroblock
j from the four luminance frame-organised sub-blocks (n=1..4)
and the four luminance field-organised sub-blocks (n=5..8)
using the intra (i.e. original) pixel values:
where
and
and Pk are the sample values in the n-th original
8*8 block.
Normalise actj:
avg_act is the average value of actj the last picture
to be encoded. On the first picture, avg_act = 400.
Obtain mquantj as:
where Qj is the reference quantization parameter
obtained in step 2. The final value of mquantj is clipped to
the range [1..31] and is used and coded as described in sections
7, 8 and 9 in either the slice or macroblock layer.