Cryptography Reference
In-Depth Information
in successive images, i.e., to encode a new frame in the video sequence, only
information which are different need to be coded and transmitted.
Although the difference between successive frames in a video sequence is
small, it will become very large when the video contains fast motion scenes.
Therefore to simply code the difference information is normally ine cient.
Alternatively, motion estimation will help to locate the MB in the reference
frame that is the best match of the current MB. This best matched MB in
the reference frame can be used as a prediction of the current MB. The dif-
ference between the MB in the current frame and its best matched MB in
reference frame is called the motion compensated difference, which forms a
low-information-content prediction-error, and is sometimes called the differ-
ence along the motion axis. Therefore greater compression ratios are possible
as we only need to send the motion vectors and the low-information-content
prediction-error. Fig. 2.10 shows the comparison of the frame difference along
the time axis and the motioned compensated difference between two consec-
utive frames in a video sequence garden. Obviously the difference along
the motion axis contains much less energy and can then be much easily com-
pressed using the DCT based Intra coding method as described in previous
section.
2.3.2 Block Matching Based Motion Estimation
Different techniques are used to estimate the motion vectors. One method
is called block matching, in which a frame is divided into a number of non-
overlapping macroblocks (MB) of size 1616. The selected MB, called current
MB, is moved around its position within a search area in the reference frame
to find the best match in the reference frame, as shown in Fig. 2.11. The
best match means the two MBs have the smallest matching error according
to certain matching criteria. A motion vector is then issued, which is defined
as the vector that pointing from the coordinate position of the current MB,
to the position of its matched MB in the reference frame. The motion vector
information forms part of the bit stream and is multiplexed with the motion
compensated difference information, which has been Intra coded.
The motion vector is obtained by minimizing a cost function measuring
the mismatch between the current MB and the MB in the reference frame. An
often used searching criterion is called the mean absolute difference (MAD),
which is defined as,
15
15
1
1616
MAD =
f (i, j)−r(i−d x ,j−d y ),
(2.8)
i=0
j=0
where,− d max
2 <d x ,d y < d ma 2 , f (i, j) is the pixel in current frame, r(i, j)is
that in the reference frame, and d max is the maximum searching displacement
horizontally or vertically. For practical algorithms, it is often not necessary
Search WWH ::




Custom Search