MPEG VIDEO CODING (Continued) |
|||||||||||||||
|
|||||||||||||||
|
MPEG Frame Coding Continued.. the position of the object 4 frames later (note that the movement has been exaggerated for clarity). The right hand section in Figure 2.4 shows the predicted position of the object after 1, 2, and 3 frames. The prediction is a linear, and based on the movement after 4 frames. The prediction for the 4th frame is transmitted as a motion vector, plus a small correction signal to give the decoder the true position of the object. For the intermediate frames there is no need to send out the motion vector for the forward step made between frames, as this can be estimated on the motion vector for movement over 4 frames. All that would be needed is any correction data, in the form of a differential DCT, for movements that were subject to acceleration or non-linear motion path.
The problem with forward prediction (i.e. the type discussed above) is that is does not cope well with a feature that was previously covered, and that is revealed as an object moves away. There is no way of predicting the properties' of the feature from forward projection. An example of this problem is shown in Figure 2.5.
In the sequence of sections from 5 successive frames in Figure 2.5 there is no way of predicting from frame 1 what the left hand side of the section will look like 4 frames later. However the forward prediction may be well able to describe the movement of the object. When the prediction is made for frame 5 left hand section it will be found that the error signal would be very large, and the data for the revealed object would be output as I data within the P frame. At a decoder (and also in the coder) this gives a buffer containing frame 1, and a second buffer containing frame 5. The revealed feature is "known" in frame 5. Therefore it is possible to look "backwards in time" from frame 5 towards frame I, identifying the object that is progressively revealed during frames 2, 3 and 4. Thus by using both forward projection (forwards in time) and backwards projection (backwards in time) it is possible to make good predictions of the intermediate revealed object and the moving object as seen in
|
||||||||||||||