Databases Reference
In-Depth Information
that would be visible in still images. On the other hand, artifacts that may not be visible in
reconstructed still images can be very annoying in reconstructed motion video sequences. For
example, consider a compression scheme that introduces a modest random amount of change
in the average intensity of the pixels in the image. Unless a reconstructed still image was being
compared side by side with the original image, this artifact may go totally unnoticed. However,
in a motion video sequence, especially one with low activity, random intensity changes can
be quite annoying. As another example, poor reproduction of edges can be a serious problem
in the compression of still images. However, if there is some temporal activity in the video
sequence, errors in the reconstruction of edges may go unnoticed.
Although a more holistic approach might lead to better compression schemes, it is more
convenient to view video as a sequence of correlated images. Most of the video compres-
sion algorithms make use of the temporal correlation to remove redundancy. The previous
reconstructed frame is used to generate a prediction for the current frame. The difference
between the prediction and the current frame, the prediction error or residual, is encoded and
transmitted to the receiver. The previous reconstructed frame is also available at the receiver.
Therefore, if the receiver knows the manner in which the prediction was performed, it can
use this information to generate the prediction values and add them to the prediction error to
generate the reconstruction. The prediction operation in video coding has to take into account
motion of the objects in the frame, which is known as motion compensation (described in the
next section).
We will also describe a number of different video compression algorithms. For the most
part, we restrict ourselves to discussions of techniques that have found their way into interna-
tional standards. Because there are a significant number of products that use proprietary video
compression algorithms, it is difficult to find or include descriptions of them.
We can classify the algorithms based on the application area. While attempts have been
made to develop standards that are “generic,” the application requirements can play a large part
in determining the features to be used and the values of parameters. When the compression
algorithm is being designed for two-way communication, it is necessary for the coding delay to
be minimal. Furthermore, compression and decompression should have about the same level
of complexity. The complexity can be unbalanced in a broadcast application, where there is
one transmitter and many receivers, and the communication is essentially one-way. In this
case, the encoder can be much more complex than the receiver. There is also more tolerance
for encoding delays. In applications where the video is to be decoded on workstations and
personal computers, the decoding complexity has to be extremely low in order for the decoder
to decode a sufficient number of images to give the illusion of motion. However, as the
encoding is generally not done in real time, the encoder can be quite complex. When the
video is to be transmitted over packet networks, the effects of packet loss have to be taken into
account when designing the compression algorithm. Thus, each application will present its
own unique requirements and demand a solution that fits those requirements.
We will assume that you are familiar with the particular image compression technique
being used. For example, when discussing transform-based video compression techniques, we
assume that you have reviewed Chapter 13 and are familiar with the descriptions of transforms
and the JPEG algorithm contained in that chapter.
