Video Compression - Introduction to Data Compression - page 645

I frame P frame B frame

Bidirectional prediction

Forward prediction

F I GU R E 19 . 14 A possible arrangement for a group of pictures.

frames. The P frames are coded using motion-compensated prediction from the last I or P
frame, whichever happens to be closest. Generally, the compression efficiency of P frames is
substantially higher than I frames. The I and P frames are sometimes called anchor frames,
for reasons that will become obvious.
To compensate for the reduction in the amount of compression due to the frequent use
of I frames, the MPEG standard introduced B frames. The B frames achieve a high level of
compression by using motion-compensated prediction from the most recent anchor frame and
the closest future anchor frame. By using both past and future frames for prediction, generally
we can get better compression than if we only used prediction based on the past. For example,
consider a video sequence in which there is a sudden change between one frame and the next.
This is a common occurrence in TV advertisements. In this situation, prediction based on
the past frames may be useless. However, predictions based on future frames would have a
high probability of being accurate. Note that a B frame can only be generated after the future
anchor frame has been generated. Furthermore, the B frame is not used for predicting any
other frame. This means that B frames can tolerate more error because this error will not be
propagated by the prediction process.
The different frames are organized together in a group of pictures (GOP). A GOP is the
smallest random access unit in the video sequence. The GOP structure is set up as a trade-off
between the high compression efficiency of motion-compensated coding and the fast picture
acquisition capability of periodic intra-only processing. As might be expected, a GOP has to
contain at least one I frame. Furthermore, the first I frame in a GOP is either the first frame of
the GOP, or is preceded by B frames that use motion-compensated prediction only from this I
frame. A possible GOP is shown in Figure 19.14 .
Because of the reliance of the B frame on future anchor frames, there are two different
sequence orders. The display order is the sequence in which the video sequence is displayed
to the user. A typical display order is shown in Table 19.5 . Let us see how this sequence was
generated. The first frame is an I frame, which is compressed without reference to any previous
frame. The next frame to be compressed is the fourth frame. This frame is compressed using