Motion-compensated predictive coding (MCPC) is the technique that has been found to be most successful for exploiting interframe correlations.
Figure 1 shows the basic block diagram of a MCPC video encoder.
 |
The transform, quantise, and entropy encode functions are basically the same as those employed for still image coding. The first frame in a sequence is coded in the normal way for a still image by switching the prediction frame to zero.
For subsequent frames, the input to the transform stage is the difference between the input frame and the prediction frame, based on the previous decoded frame. This difference frame is usually known as the prediction error frame.
The purpose of employing predictions is to reduce the energy of the prediction error frames so that they have lower entropy after transformation and can therefore be coded with a lower bit rate.
If there is motion in the sequence, the prediction error energy may be significantly reduced by motion compensation. This allows regions in the prediction frame to be generated from shifted regions from the previous decoded frame. Each shift is defined by a motion vector which is transmitted to the decoder in addition to the coded transform coefficients. The motion vectors are usually entropy coded to minimise the extra bit rate needed to do this.
The multiplexer combines the various types of coded information into a single serial bit stream, and the buffer smoothes out the fluctuations in bit rate caused by varying motion in the sequence and by scene changes. The controller adjusts coding parameters (such as the quantiser step size) in order to maintain the buffer at approximately half-full, and hence it keeps the mean bit rate of the encoder equal to that of the channel.
Decoded frames are produced in the encoder, which are identical to those generated in the decoder. The decoder comprises a buffer, de-multiplexer, and entropy decoder to invert the operations of the equivalent encoder blocks, and then the decoded frames are produced by the part of the encoder loop comprising the inverse quantiser, inverse transform, adder, frame store, motion compensator and switch.
H.261 is a CCITT standard for video encoding for video-phone and video conferencing applications. Video is much more important in a multi-speaker conferencing environment than in simple one-to-one conversations. H.261 employs coders of the form shown in Figure 1 to achieve reasonable quality head-and-shoulder images at rates down to 64 kb/s (one ISDN channel). A demonstration of H.261 coding at 64 and 32 kb/s will be shown.
A development of this, H.263, is now in current use, to allow the bit rate to be reduced down to about 20 kb/s, without too much loss of quality, for modems and mobile channels. This uses some of the more advanced motion methods from MPEG (see later), and it also forms the basis for baseline MPEG-4 coders.
