GB2240235A - DATV coding of high definition television signals - Google Patents

Abstract

In a DAVT HD-MAC system using 80ms, 40ms and 20ms coding branches, in the 80ms branch, instead of taking one quarter of the samples from each of four fields or all the samples from one field in four, half the samples are taken from half the fields, considered in pairs. The information is described as derived from a sequential image co-timed with one field and is transmitted in two adjacent fields. A "spot-wobble" technique due to estimated motion vectors may be employed. The new signal is compatible with first-generation DATV receivers while providing an improved picture on second-generation DATV receivers. <IMAGE>

Description

TELEVISION SYSTEMS This invention relates to television systems in which a television signal is transmitted which comprises picture information at a normal level of definition and, in a non-active-picture part of the signal, additional coded information representative of further picture detail. A normal definition receiver receives and displays the picture at the normal level of definition. A high definition receiver receives the picture at the normal level of definition, also receives and decodes the additional information, and modifies the normal definition picture on the basis of the decoded additional information, to give a picture which is displayed at a higher definition.

Current proposals for the coding of signals in a format compatible with the MAC/packet family of standards use a number of coding branches, some of which employ compensation for motion vectors within the scene. For a background description of such proposals see a paper by Vreeswijk, F.W.P. et al. "HD-MAC coding of high definition television signals" published in Proceedings of the 1988 International Broadcasting Convention (IEE Conference Publication No. 293), pp 62-65. As described in that paper a MAC video signal is transmitted in analogue form and consists of two time-multiplexed components; chrominance and luminance respectively. Extra high definition information is added to this signal in such a way that a normal definition receiver can display the MAC signal with minimal impairments.A high definition receiver extracts the extra information and displays a high definition picture. The system described treats the picture in blocks of, say, eight samples by eight lines, and for each block compares successive fields to determine the rate of movement of the picture in the block. Three different velocity ranges are identified, defined as the 20ms, 40ms and 80ms branches, and one of three different processor branches used in dependence on the velocity determined for that block. Each branch effects a different trade-off between spatial and temporal detail. It is noted in the above paper that the 80ms and 40ms branches may include motion compensation. For this purpose motion vectors are transmitted and used in the decoder.The selection of which branch is to be used in the decoder, together with the values of the motion vectors, if appropriate, is signalled to the receiver using DATV (digitally assisted television) information in the field blanking interval.

The system described in the aforementioned paper has been termed the P3 system. This application is concerned with improvements in the system.

We have appreciated that the P3 coding process introduces impairments in the decoded HDTV picture which are likely to become an increasing problem as cameras and displays improve. However, it is desirable that any improvements to P3 should be introduced without causing significant additional impairments to pictures viewed on MAC or first-generation HD-MAC receivers.

We have also appreciated that unexpected benefits may be obtained by the use of motion compensation in the 80ms branch.

We attribute this directly to the additional resolution provided in slowly moving areas, and the lack of switching artifacts as objects start to move.

In accordance with this invention, which is defined in the appended claims, we propose that in the 80ms branch samples are taken from fields separated by 40ms, two successive sets of samples being used to reconstruct images at the decoder. At the decoder, the necessary decoding can be achieved by the use of filters of the same type as normally used for interlace-to-sequential scan conversion.

In the proposed system, alternate fields are subsampled and the samples thus-obtained used to reconstruct four fields, whereas the P3 coding system either subsamples alternate fields and reconstructs two fields with each set of samples, or subsamples each or every fourth field to reconstruct groups of four successive fields.

This improvement can be implemented by using bit-rate reduction techniques on the DATV data, as described in our UK Patent Application No. 90 1F011. of even date entitled "Improvements in Television Systems" (agent's reference 30898), together with minor changes to the transmitted analogue signal.

The invention will be described in more detail by way of example with reference to the drawings, in which: Figure 1 is a block diagram of the P3 HD-MAC coding system taken from the above-mentioned IEE paper; Figure 2 is a diagram illustrating the sampling structure and interpolation processes in one example of a "compatiblen 80ms motion-compensated branch; Figure 3 is a diagram illustrating the pass-bands of the 80ms branch in the presence of non-compensated motion for two different sampling options; and Figure 4 is a diagram illustrating the use of perturbations to overcome loss of resolution.

Referring to Figure 1, this shows a P3 HD-MAC system as described in the above-mentioned paper. Reference should be made to the paper for a fuller description thereof. Briefly, however, at the encoder 10 the output of an HD-camera 12 is applied to three coding branches 14, 16 and 18 based on one, two or four fields of the 1250 line 50 field per second interlaced signal from the camera. A movement processor 20 senses whether movement occurs in the picture and controls the 40ms and 80ms branches accordingly. A switch SW1 selects the appropriate channel output in dependence upon the output of the movement processor.

The encoder output passes in the form of a 625 line 50 field per second interlaced picture through a conventional transmission channel 22 to a decoder 24. The decoder also uses three branches 26, 28, 30. A switch SW2 selects the appropriate branch output for display on a monitor 32.

Information on the coding branch in use and on the movement vector is transmitted by a bit-rate reduction (BRR) encoder 34 as the DATV signal through the channel 22. A BRR decoder 36 decodes this information and controls the decoder in accordance therewith.

As shown, the 20ms branch is split into three sub-branches controlled by a frequency analyser 38.

In order to add motion compensation to the 80ms branch, a method is required that ensures that the pictures produced by both a simple MAC receiver and by a first-generation HD-MAC receiver are not unduly impaired. A possible way of achieving this is described below; the sampling and interpolation processes proposed are illustrated in Figure 2.

The HD image is sub-sampled at the same spatial coordinates as used for the existing 80ms branch. This is shown in Figure 2(a), where the numbers indicate which field is sampled and the field in which the sample is transmitted. However, in accordance with this invention the four phases of samples are divided into two pairs; each pair of samples being derived from a sequential image co-timed with an incoming field 1. Thus information is transmitted derived from half the sampling positions on each of only two of the input fields, rather than one quarter of the samples on each field or all the samples of every fourth field, as has been proposed previously.

One way in which this can be done is illustrated in Figure 2(b), which shows a preferred sampling structure for the 80ms motion compensated (MC) branch modified from Figure 2(a).

There is then some slight sample repositioning, as shown in Figure 2(c), in particular vertical sample re-ordering, and then the samples closely resemble and are compatible with those from the 40ms branch, and the same motion-compensated compatibility improvement technique is used.

In order to make branch decisions, the encoder forms the signal from the decoded 80ms-MC (motion compensated) branch, and uses it in the a Posteriori decision-making process in place of the signal from the existing stationary 80ms branch. It may be worthwhile to limit the velocity range of this branch to less than that of the 40ms branch in order to reduce compatibility problems and simplify decoder hardware; if the estimated motion vector for a given block was outside this range, the coder would be forced to select either the 20ms or 40ms branches.

The branch selection information can be sent separately from the existing P3 data, using the spare channel capacity.

Alternatively, if a menu system is used as a part of the P3 DATV bit-rate reduction process, this information can be sent together with the P3 data. The motion vectors can be the same as those used for the 40ms branch, and are transmitted as a part of the P3 data, since the first-generation receivers will decode 80ms-MC blocks as if they were coded via the 40ms branch. Either of these alternatives could form part of a system to implement smaller decision blocks.

A second-generation decoder uses the last four fields of transmitted samples to interpolate sequential images co-timed with each field 1, using a motion-compensated filter. This provides discrete processing over 40ms, but continuous processing from one 40ms period to the next. In "continuous" processing, a proportion (1/n) of the sites in each incoming field are subsampled and the decoder reconstructs each image from the n most recently transmitted fields; in "discrete" processing all sample sites from every nth incoming field are sampled and the decoder uses one (or two) set(s) of such samples to reconstruct the immediately preceding and following fields.

If the sampling arrangement shown in-Figure 2(b) is used, the filter resembles a motion-compensated interlace-to-sequential converter operating on diagonal lines.

Even fields are then generated from the interpolated sequential images by motion-compensated temporal interpolation, in the same way as in the present 40ms branch decoder. One motion-compensated filter can be used to carry out both of these processes; this filter is also used to generate signals for the stationary 80ms branch, which is essentially a sub-set of the motion-compensated branch. The decoding of the 80ms MC branch in a second generation decoder is illustrated in Figure 2(d).

The samples obtained by the MC interpolation filter, shown by small circles, are those that are filtered in a manner similar to that in interlace-to-sequential conversion.

This decoding process is able to generate images with a spatial frequency content equal to that of the stationary 80ms branch for a range of velocities; however at some "critical" speeds of an odd number of pixels per field period, the potential resolution is reduced. Figure 3(a) illustrates the worst-case situation for the sampling pattern shown in Figure 2(b); the loss in resolution is asymmetric, but full horizontal resolution can always be maintained. Figure 3(a) shows on the left the passband of the 80ms MC branch with sample phases 1 and 2 taken from the same field as shown on the right. It maintains full horizontal resolution but has asymmetric diagonal resolution in the presence of non-compensated motion.The area marked out by the line X defines a possible passband for the MC branch for motion speeds of one or an odd number of pixels per field period; this is the worst case for horizontal or vertical motion. The diagonal area marked out by the line Y indicates the 80ms passband for motion speeds of zero or an even number of pixels per field period.

Another option is to co-time the sampling of sites 1 and 3 instead of 1 and 2 in Figure 2(a); this results in isotropic resolution loss in the presence of non-compensated motion, although this implies loss of horizontal resolution. This is illustrated in Figure 3(b) which shows the passband of the 80ms MC branch with sample phases 1 and 3 taken from the same field as shown on the right. The square area Z shows the possible 80ms MC passband at motion speeds at the worst case of one or an odd number of pixels per field period in the diagonal direction.

Other sampling patterns may exist that offer better compromises. However, the loss of resolution could be lessened by introducing small perturbations to the sampling structure, using a principle similar to "spot wobble". This would not significantly affect encoder or decoder complexity, as it requires no more than slight changes to filter coefficients as a function of the motion speed over the past few frames.

The principle of "spot wobble" is explained by reference to Figure 4. This figure shows a line of samples on four successive fields and the position of an object moving at the critical speed of one pixel per field period horizontally. The sample sites are displaced by two pixels in alternate odd fields in order to ensure that the object is sampled at the correct spatial sites. The requirement for such a displacement is determined from the measured motion vector; as this is known to the decoder as well as the encoder, the inverse distortion can be applied during the reconstruction process. This results in all required sites on the moving object being sampled in two successive 40ms periods. Without "spot wobble" the same sites would be sampled in successive images, so a complete set of samples would not be available.

The additional impairments introduced to pictures from first-generation decoders should not be serious. They take the form of slight additional dot-patterning in areas transmitted using the 80ms branch for other than near-zero motion speeds, and may look similar to the effect of the non-linear post-filter used in the 40ms branch decoder. No additional significant motion impairments are introduced, with the possible exception of a very slight "wobble" in moving areas if the spot-wobble technique mentioned above is employed. The effect on the MAC-compatible picture is similar, although the compatibility filtering techniques limit the visibility of any additional dot-patterning.

Claims (5)

1. An encoder providing a television signal which comprises picture information at a normal level of definition and, in a non-active-picture part of the signal, further information relating to each of a plurality of regions in the picture and including an indication of which of a plurality of coding modes and/or motion vectors is applicable to each region characterised in that one of the coding modes derives information for four successive transmitted fields from samples of alternate fields, the said samples and the further information enabling the reconstruction of four fields at a decoder.

2. An encoder according to claim 1, in which in the said mode indication is derived from a sequential image co-timed with one field and is transmitted in two adjacent fields.

3. An encoder according to claim 1, in which samples taken in the said one of the modes are spatially positioned prior to transmission to resemble those generated by another coding mode, thereby allowing this other coding mode to be used in decoders not equipped to deal with the additional mode.

4. An encoder according to claim 1, in which the sampling sites in alternate sampled images are displaced from their conventional positions in accordance with the estimated motion vector, in order to ensure that objects remain correctly sampled when moving at speeds which would otherwise cause aliasing.

5. A decoder for receiving a signal provided by the encoder of any preceding claim and displaying a high definition-picture therefrom by making use of the said further information.