VIDEO SIGNAL PROCESSING
This invention concerns the compression of video signals, and in particular aspects, the manipulation of motion vectors accompanying an associated video signal through a video process.
Motion vectors can be used in a wide range of video processes such as data compression, noise reduction and standards conversion. The extraction of accurate vectors from a video signal is a complex and difficult process and it is often more efficient to carry the vectors in a parallel path associated with the video signal than to re-create them from the video whenever they are required. A particular problem with video data compression systems is that the subjective quality can become seriously degraded if several compression and decompression processes are cascaded. A known technique for alleviating this degradation is to carry the motion vectors and other parameters used in the first compression process with the video and to use the same vectors in subsequent compression processes (for example, see SMPTE standards 327M and 319M: MPEG re-encoά' g data set and its transport mechanism).
However, in certain circumstances, arising for example in recording, transmission or post-production, an MPEG-coded video signal may be required to pass through a low bandwidth or non-MPEG pathway. The processes involved may have limited data capacity which may be exceeded by signals having accompanying previous coding parameters.
It is therefore an object of the invention to address these problems and, in certain aspects, to reduce the data content of parameters such as motion vectors accompanying a video signal.
Accordingly, the invention consists in one aspect in a method of video signal processing comprising the steps of: receiving a decoded video signal and an associated coding parameter; using a deterministic process to provide a first prediction of the coding parameter; comparing the coding parameter with the first prediction of the coding parameter to form a comparison signal; replacing the associated coding parameter with the comparison signal; transporting the video signal and associated comparison signal along a video pathway; using the
deterministic process to provide a second prediction of the coding parameter; reconstructing the coding parameter by comparing the second prediction of the coding parameter with the comparison signal; and replacing the comparison signal with the reconstructed coding parameter.
The invention will now be described by way of example with reference to the accompanying drawings, in which:
Figure 1 is a diagram illustrating the processing of a decoded video signal; and
Figure 2 is a diagram illustrating the processing of motion vectors according to an embodiment of the invention.
Figure 1 illustrates an example of a problem to which the invention is addressed. An MPEG-coded video signal (101) is required to pass through a baseband video (or other non-MPEG) process (103). The process may include recording, transmission or post-production of the video.
At the output of the process the video is required to be re-encoded to the
MPEG format and, in order to reduce the artefacts introduced by this multiple coding, the motion vectors which were used to code the video signal (101) are extracted by the MPEG decoder (102), passed through the process (103) and used to re-code the video in the MPEG coder (104) to create a processed MPEG signal (105).
The motion vectors included in the signal (101) may be highly accurate and have been derived from an "upstream" motion estimator (which could be incorporated in the previous MPEG coder). The process (103) may have limited data capacity in its motion vector channel and, if so, it may be necessary to reduce the data content of the motion vectors so that they can fit into this limited capacity.
The data reduction, and subsequent data expansion, of the vectors according to an embodiment of the invention will now be described with reference to Figure 2.
A video signal (201) is accompanied by motion vectors (202). Data reduction of the motion vectors is achieved by making a prediction (204) of the vectors and subtracting (205) the predicted vectors from the actual vectors to obtain a motion vector prediction error signal (206). This error signal accompanies the video signal through the process (203) and, when motion vectors are required for subsequent processing, they are derived from it as will be explained subsequently.
The prediction is made by analysing the video signal (201) in a motion estimator (207). This motion estimator need not be highly accurate as its output vectors will not be used directly. However, if the estimation process is deterministic, a particular sequence of input video frames will always result in the same output vectors, thus allowing reconstruction of the vectors after the process (203), as described below.
The input motion vectors (202) will differ only slightly from the predicted vectors (204) and so the prediction error signal (206) will usually be small. The prediction error is coded for onward transmission in the block (208). This coder has a lower output data rate than the input vector signal (202); this data reduction can be achieved by variable length (Huffman) coding, though other methods will be apparent to those skilled in the art.
The video and the reduced-data-rate vector prediction error signal (209) are associated with each other in the process (203). Where subsequent motion- compensated processing of the video is required (for example, MPEG encoding) the original, full-data-rate motion vectors are reconstructed as follows.
The video signal (210) is applied to a motion estimator (211), which has the same deterministic input to output relationship as that (207) used to produce the motion vector error signal (206). As the video signal (210) is the same as the input video (201 ), the output (212) from this second motion estimator should be the same as the original motion vector prediction (204).
The low-bit-rate vector prediction error signal (213) at the output of the process (203) is restored to the original bit rate in the block (214); this is achieved by the reverse of the process in block (208). This full data rate error signal is added (215) to the predicted vector signal (212) from the motion estimator (211) so as to obtain a motion vector signal (216) which is equivalent to the original signal (202). In order to retain synchronisation between the video signal and the motion vector signal, compensating vector delays (217, 218), and compensating video delays (219, 220) are included as shown in Figure 2.
Thus, in this manner, the vector information accompanying a video signal is reduced to allow it to pass through a process with the video, providing the information required to enable lossless re-encoding at a later stage. It is therefore possible to employ vectors for lossless concatenation in numerous processes in which it would otherwise be problematic.
It will be appreciated by those skilled in the art that the invention has been described by way of example only, and that a wide variety of alternative approaches may be adopted without departing from the scope of the invention. In particular, though the above examples describe the use of motion vectors, various other parameters reflecting previous coding decisions, such as the coding modes or quantisation levels employed, may be estimated and employed in compression in similar vein.