Databases Reference
In-Depth Information
T A B L E 19 . 5
A typical sequence of frames in
display order.
IBBPBBPBBPBBI
123456789 0 1 2 3
motion-compensated prediction from the first frame. Then we compress frame 2, which is
compressed using motion-compensated prediction from frame 1 and frame 4. The third frame
is also compressed using motion-compensated prediction from the first and fourth frames.
The next frame to be compressed is frame 7, which uses motion-compensated prediction from
frame 4. This is followed by frames 5 and 6, which are compressed using motion-compensated
predictions from frames 4 and 7. Thus, there is a processing order that is quite different from
the display order. The MPEG document calls this the
bitstream order
. The bitstream order for
the sequence shown in Table
19.5
is given in Table
19.6
. In terms of the bitstream order, the
first frame in a GOP is always the
I
frame.
As we can see, unlike the ITU-T H.261 algorithm, the frame being predicted and the
frame upon which the prediction is based are not necessarily adjacent. In fact, the number of
frames between the frame being encoded and the frame upon which the prediction is based
is variable. When searching for the best matching block in a neighboring frame, the region
of search depends on assumptions about the amount of motion. More motion will lead to
larger search areas than a small amount of motion. When the frame being predicted is always
adjacent to the frame upon which the prediction is based, we can fix the search area based on
our assumption about the amount of motion. When the number of frames between the frame
being encoded and the prediction frame is variable, we make the search area a function of the
distance between the two frames. While the MPEG standard does not specify the method used
for motion compensation, it does recommend using a search area that grows with the distance
between the frame being coded and the frame being used for prediction.
Once motion compensation has been performed, the block of prediction errors is trans-
formed using the DCT and quantized, and the quantization labels are encoded. This procedure
is the same as that recommended in the JPEG standard and is described in Chapter 13. The
quantization tables used for the different frames are different and can be changed during the
encoding process.
Rate control in the MPEG standard can be performed at the sequence level or at the level
of individual frames. At the sequence level, any reduction in bit rate first occurs with the
B
frames because they are not essential for the encoding of other frames. At the level of the
individual frames, rate control takes place in two steps. First, as in the case of the H.261
T A B L E 19 . 6
A typical sequence of frames in
bitstream order.
IPBBPBBPBBI BB
1423756 089 3 1 2
