GB2190813A - Improvements in predictors for video signals - Google Patents

Abstract

A prediction signal is substracted 14 from a video input signal to form a differential signal for transmission via a coder 46. The differential signal quantized 16 is added 20 to the prediction signal to form a predictor input signal from which the prediction signal is formed by means of pixel and line delays 24, 32, 36, 34; multipliers 26, 42, 44 which multiply by weighing coefficients C1, C2 etc. and adders 40, 38, 28 which sum the partial products. The various delays enable the prediction signal to be formed, with enhanced accuracy, from a cluster of samples (at least six and preferably more) taken from a plurality of adjacent lines and, optionally, from more than one consecutive field or picture. Limiters 22 and 30 ensure that neither the predictor input signal nor the prediction signal can go out of range. Advantageous sets of weighing coefficients are disclosed. <IMAGE>

Description

SPECIFICATION Improvements in Predictors for Video Signals This invention relates to a predictor for use with video signals, and useful in particular in differential pulse code modulation (DPCM) systems for the bit rate reduction of digital television signals.

In DPCM coding a prediction is made for each incoming sample of the input signal and the difference between the input and prediction is quantised for transmission to the DPCM decoder.

The prediction is formed as a weighted sum of contributions from previously-decoded sample values such that both the coder and the decoder can form the same prediction values. For a given degree of bit rate reduction the quality of the decoded picture is optimised by optimising the accuracy of the prediction. The weighted sum is formed from a plurality of previously decoded sample values multiplied by respective weighing coefficients.

Existing DPCM systems use simple predictors which employ only a few coefficients or use a combination of two or more simple predictors which are selected adaptively.

Predictors can be used in other circumstances where an estimate of the next sample value of a video signal is required. Examples of this are in error concealment systems or in noise reduction systems.

The predictor according to the present invention is defined in the appended claims. The invention will be described in more detail, by way of example, with reference to the accompanying drawings, in which: Figure 1 is a block circuit diagram of a DPCM coder, Figure 2 is a diagram showing sampling positions on a vertical-horizontal grid of a composite PAL signal, and Figure 3 is a diagram showing sampling positions on a vertical-horizontal grid for the luminance (or chrominance) component of a component coded video signal sampled with a line locked sampling pattern.

An example of an arrangement for the construction of a DPCM coder embodying the invention is shown in Fig. 1. The DPCM coder has an input 12 which receives a video input signal. This is applied to the non-inverting input of a subtractor 14 to the output of which is connected a nonlinear quantiser 16. The quantiser output is connected to an adder 20 the output of which is applied through a range limiter 22 to a pixel delay 24 providing a delay of one pixel (picture element).

A multiplier 26 multiplies the output of the pixel delay 24 by a coefficient C1 and applies the product to an adder 28. The output of the adder 28 is fed through another range limiter 30 both to the inverting input of subtractor 14 and to the second input of the adder 20.

As thus-far described the circuit includes a standard DPCM loop (subject to the addition of the range limiters for purposes described below). An input sample is compared with the previous transmitted sample in subtractor 14 and the difference quantised in the non-iinear quantiser 16.

The quantiser output is added to the previous prediction emanating from circuit 30, delayed by one sample period or pixel, and then applied to the subtractor to be compared with the next incoming sample. Thus the output of the range limiter 30 constitutes a prediction for the next sample value, in this case the prediction being an unsophisticated one based simply on a summation of preceding sample differences.

The prediction can however be made more sophisticated by using a weighted sum of previous terms as indicated by the additional circuitry shown in Fig. 1. In the particular example illustrated there is a line delay 32 and various further pixel delays 34, 36 etc. inserted in series with adders 38, 40 etc. The adders are fed by multipliers (or attenuators) 42, 44 etc. multiplying by coefficients C2, C3 etc, so that samples from the previous line can be employed. It will be appreciated that by the addition of further delays, multipliers and adders, additional terms from the same line of the video signal and terms from previous lines or from interlaced lines from the previous field can be taken into consideration. Samples from previous pictures (frames) of the interlaced video can be used if delays totalling a picture delay are incorporated.

Finally it should be mentioned that the output of subtractor 14 is transmitted as the output of the coder and to this end is applied through a transmission code generator 46 and conveniently a further pixel delay 48 to the output 50.

Because the quantiser 16 is a non-linear quantiser, it is possible for it to produce an output which, when combined in adder 20, would exceed the capacity of the processing circuitry. If this happens sample values can be erroneously interpreted, eg. in an 8-bit binary system a sample of value 257 would be interpreted as having a value of 1. To stop this happening the range limiter 22 is included which stops the signal going out of range. A similar effect can be achieved at the output of adder 28 which represents the sum of several coefficient terms, and so range limiter 30 is also included.

Fig. 1 shows a coder but, as is very well known, a DPCM coder effectively includes the circuitry of a DPCM decoder. Thus a decoder is not separately described but is constitued by the circuitry to the right of the quantiser in Fig. 1.

The techniques described below are suitable for non-adaptive DPCM coding in which the prediction is formed from the weighted contributions from several previous samples, e.g. six or more. They provide a system which is stable in operation and is relatively insensitive to transmission errors.

Fig. 2 illustrates the manner of definition of the samples in a composite PAL signal with a sample rate of twice the colour sub-carrier frequency, 2fisc. As will be seen, because the sampling pattern involves a line to line offset, a 4fsc grid is used to accomodate all the sampling points. Each sample can be defined as S(i,j,k) where S(O,O,O) is the sample to be predicted, the "current sample", and: i indicates how many sample positions the sample is to the left of the current sample; j indicates how many interlaced lines the sample is above the line containing the current sample; and k indicates how may complete pictures (or frames) the sample is prior to the current sample.

The first examples are all concerned with samples taken from a single picture, i.e. k is zero.

Using this notation, we have found that predictors having the coefficients C indicated in Tables 1 and 2 below are particularly suitable for use with composite PAL signals. The predictor of Table 1 has more components than that of Table 2 and, on average, can be used to give a greater degree of bit rate reduction for a given picture quality.

Each table shows for each sample point used the preferred coefficient value, and the tolerance (plus or minus) associated with that value. It will be seen that each coefficient has a fairly broad optimum range of values. However the algebraic sum of all the coefficients used should be a little less than unity and is preferably in the range between 0.85 and unity. However, there may also be a tolerance in the overall gain of the predictor. For example, the predictor is essentially the same if all coefficients are multiplied by a constant factor between 0.85 and 1.05.

As indicated in the last line of each Table, small amounts of data from other samples than those specifically listed may be permitted.

Fig. 3 and Tables 3 and 4 give corresponding information for a component coded PAL video signal sampled at the colour subcarrier frequency f,, with a line locked sampling system. Here a regular grid results rather than the offset grid of Fig. 2. A similar form of definition of i, j, k is used. Table 3 illustrates a first example using samples from one frame only while Table 4 shows an example taking contributions also from the previous frame (i.e k= 1).

Again small contributions can be taken from samples which are not illustrated. The total coefficient values should be in the same range, 0.85 to unity, as for the previous examples.

Tables 5 and 6 represent preferred coefficients for the luminance and chrominance components respectively in a component coded signal. The basic structure for each component is again as shown in Fig. 3 but higher and lower sampling frequencies are used for the luminance and chrominance components respectively, e.g. the luminance sampling frequency is twice the chrominance sampling frequency. Having said this, there is no reason in principle why either set of coefficients should not be used with either the luminance or chrominance component.

The examples illustrated by the tables are found to provide subjectively improved transmission performance for a given data rate or alternatively permit a reduction in data rate for the same picture quality.

While the predictors have been described in the context of a DPCM system, where they are particularly suitable, they could find utility in other applications such as in error concealment or in noise reduction.

TABLE 1 Coordinates according Predictor coefficients C to Fig. 2 Tolerance i fl k Value (Plus or Minus) 2 0 0 1/8 1/16 4 0 0 1/2 1/8 -1 2 0 1/4 1/16 1 2 0 1/16 1/16 3 2 0 -1/8 1/16 -2 4 0 5/32 1/16 0 -1 0 3/8 1/16 2 -1 0 1/16 1/16 4 -1 0 -1/4 1/16 -1 1 0 3/16 1/16 1 1 0 3/16 1/16 3 1 0 -5/32 1/16 5 1 0 -3/32 1/16 -2 3 0 -5/16 1/16 All other i,1, k 0 1/16 TABLE 2 Coordinates according Predictor coefficients C to Fig. 2 Tolerance i j k Value (Plus or Minus) n 2 0 1/8 1/16 o 4 0 1/2 1/8 -1 2 0 5/16 1/16 1 2 0 3/32 1/16 3 2 0 -3/16 1/16 O -1 0 3/8 1/16 2 -1 0 1/4 1/16 4 -1 0 -1/2 1/8 All other i, .1, k 0 1/16 TABLE 3 Coordinates according Predictor coefficients C to Fig. 3 Tolerance i 1 k Value (Plus or Minus) 1 0 0 0.723 0.3 2 0 0 -0.133 0.1 3 0 0 0.181 0.1 -1 2 0 0.106 0.1 0 2 0 0.123 0.1 1 2 0 -0.044 0.05 2 2 0 0.038 0.05 3 2 0 -0.066 0.05 0 -1 0 0.452 0.1 1 -1 0 -0.291 0.1 2 -1 0 0.041 0.05 3 -1 0 -0.093 0.05 -2 1 0 -0.025 0.05 -1 1 0 -0.043 0.1 0 1 0 0.312 0.1 1 1 0 -0.283 0.1 All other i, .1, k 0.000 0.05 TABLE 4 Coordinates according Predictor coefficients C to Fig. 3 Tolerance i j k Value (Plus or Minus) 1 0 0 0.995 0.3 2 0 0 -0.455 0.2 3 0 0 0.313 0.2 4 0 0 -0.116 0.1 5 0 0 -0.066 0.05 -1 2 0 0.000 0.05 0 2 0 0.339 0.1 1 2 0 -0.243 0.1 2 2 0 0.115 0.1 3 2 0 -0.065 0.052 0 -1 0 0.226 0.1 1 -1 0 -0.171 0.1 2 -1 0 0.000 0.05 3 -1 0 0.000 0.05 -2 1 0 n.000 0.05 -1 1 0 0.000 0.05 0 1 0 0.190 0.1 1 1 0 -0.190 0.1 0 0 1 0.382 0.1 1 0 1 -0.363 0.1 2 0 1 0.167 0.1 3 0 1 -0.093 0.1 0 1 1 -0.261 0.1 1 1 1 0.169 0.1 2 1 1 -0.009 0.05 All other i, 1, k 0.000 0.05 TABLE 5 Coordinates according Predictor coefficients C to Fig. 3 Tolerance i .1 k Value (Plus or Minus) 1 0 0 1.000 0.20 2 0 0 -0.437 0.10 3 0 0 0.235 0.10 0 2 0 0.327 0.10 1 2 0 -0.195 0.07 0 -1 0 0.201 0.07 1 -1 0 -0.158 0.07 0 1 0 0.172 0.07 1 1 0 -0.172 0.07 0 0 1 0.460 0.10 1 0 1 -0.497 0.10 2 0 1 0.277 0.10 3 0 1 -0.149 0.07 0 2 1 -0.291 0.10 1 2 1 0.198 0.07 All other i, i, k 0.000 0.07 TABLE 6 Coordinates according Predictor coefficients C to Fig. 3 Tolerance i .i k Value (Plus or Minus) 1 0 0 0.500 0.10 -1 2 0 0.194 0.10 0 2 . 0 0.272 0.10 n 0 1 0.655 0.15 1 0 1 -0.317 0.10 -1 2 1 -0.137 0.07 0 2 1 -0.191 0.10 All other i, i, k, 0.000 0.07

Claims (20)

1. A predictor for use with video signals, comprising means for multiplying a plurality of differently delayed video signal samples by corresponding weighting coefficients and means for scanning the resulting product signals to form a prediction signal, characterised in that the delayed video signal samples include samples delayed relative to other samples by one or more line periods and for one or more field periods and/or one or more picture periods.

2. A predictor according to claim 1, characterised in that the plurality of components includes at least six components.

3. A predictor according to claim 2, characterised in that the delayed samples also include samples delayed relative to other samples by one or more picture element periods.

4. A predictor according to claim 3, characterised in that the samples consist of a cluster of samples distributed over a plurality of adjacent video lines, with a plurality of samples taken from at least some of the lines.

5. A predictor according to claim 4, suitable for use with a PAL signal, and substantially in accordance with Fig. 2 of the drawings and Table 1.

6. A predictor according to claim 4, suitable for use with a PAL signal and substantially in accordance with Fig. 2 of the drawings and Table 2.

7. A predictor according to claim 4 and substantially in accordance with Fig. 3 of the drawings and Table 3.

8. A predictor according to claim 3, characterised in that the samples consist of a cluster of samples distributed over a plurality of adjacent video lines from two consecutive pictures, with a plurality of samples taken from at least some of the lines.

9. A predictor according to claim 8 and substantially in accordance with Fig. 3 of the drawings and Table 4.

10. A predictor according to claim 8 and substantially in accordance with Fig. 3 of the drawings and Table 5.

11. A predictor according to claim 8 and substantially in accordance with Fig. 3 of the drawings and Table 6.

12. A predictor according to any preceding claim, characterised in that prediction is effected separately for separate luminance and chrominance components, utilizing lower and higher sampling frequencies for these two components respectively.

13. A predictor according to claim 12, characterised in that luminance and chrominance prediction are in accordance with claims 10 and 11 respectively.

14. A predictor according to any preceding claim, characterised in that the algebraic sum of the weighting coefficients lies in the range 0.85 to 1.00.

15. A predictor according to any preceding claim, characterised in that the prediction signal is range limited.

16. A predictor according to any preceding claim, characterised in that each weighting coeffi cient is assigned a corresponding tolerance of permissable variation thereof.

17. A predictor according to claim 16, characterised in that the samples form a cluster in horizontal/vertical/time space and that the tolerances are the smaller the more distant the samples are from the centre of the cluster.

18. A differential pulse code modulation system comprising a predictor according to any of claims 1 to 17, means for subtracting the prediction signal from an input video signal to form a differential signal, and means for adding the prediction signal to the quantized differential signal to form a predictor input signal.

19. A system according to claim 18, characterised in that the predictor input signal is range limited.

20. A system according to claim 18 and substantially as illustrated in Fig. 1 of the accompanying drawings.