GB2215160A

GB2215160A - Picture coder/decoder with motion-compensated adaptive predictor

Info

Publication number: GB2215160A
Application number: GB8803595A
Authority: GB
Inventors: Michael James Knee; Nicholas Dominic Wells
Original assignee: British Broadcasting Corp
Current assignee: British Broadcasting Corp
Priority date: 1988-02-17
Filing date: 1988-02-17
Publication date: 1989-09-13
Anticipated expiration: 2008-02-17
Also published as: GB2215160B; GB8803595D0

Abstract

A coder for a television signal I subtracts a predicted signal P to produce an error signal E which is quantized to produce the coded signal E'. An adaptive predictor 15 provides the predicted signal P in response to a locally decoded signal L. The predictor coefficients are provided by a coefficient calculator 16 responsive to sample values of the locally decoded signal L from a plurality of fields stored in a memory 17. The prediction is thus effective in both the temporal and spatial domains and effects inherent motion compensation. The coder is preferably a hybrid transform coder with a transform 11 operating on the error signal and an inverse transform 13 operating on the quantized error signal. <IMAGE>

Description

PICTURE CODER/DECODER WITH MOTION-COMPENSATED ADAPTIVE PREDICTOR The invention relates to a motion-compensated adaptive predictor for use in picture coding systems that involve temporal as well as spatial, processing of the picture.

In particular, the predictor could be used to provide for an additional predictor in a DPCM differential pulse code modulation system based on switched prediction. The new predictor would be particularly suitable for detailed, moving areas of the picture.

Other predictors would be made suitable for other areas; for example, one could have a simple spatial predictor for plain areas and a pure interframe predictor for other stationary areas.

Alternatively, the new predictor could be used to provide a motion-compensated interfield or interframe prediction for a hybrid transform coding system, as described below.

In the temporal element, based on the invention, of a threedimensional picture coding system such as adaptive DPCM or hybrid transform coding, the picture is divided into between 1 and, say, 100 different areas, either of equal size or as determined by the changing picture statistics. For each area, the coefficients of a predictor based on locally decoded or input samples in previous fields of the picture signal are calculated by one of several methods, examples of which are given below. The calculation process may have an element which ensures that the resulting predictor is stable in the presence of transmission errors, for example as described in Patent Application No. 88 00360. For each picture area, new coefficients are calculated either on a regular basis (for example, once per field) or as determined by the changing predictor statistics.

The advantage of the invention in each case is that it provides, especially in moving areas of the picture, a more accurate prediction than do conventional fixed predictors such as the pure interframe predictor that might be used in a hybrid transform coding system. This enables the resulting prediction error signal to be encoded at a lower bit rate or to higher accuracy (the latter giving a higher decoded picture quality).

Embodiments of the invention will now be described, by way of example, with reference to the accompanying drawings, in which: Fig. 1 is a block diagram of a DPCM coder, and Fig. 2 is a block diagram of an alternative DPCM coder.

Both embodiments use a hybrid transform coding system in which the prediction error signal is encoded using spatial transform coding based on, for example, the discrete cosine transform applied to small blocks of the picture area. The predictor could also be used in an adaptive hybrid transform coding system.

In Fig. 1 the input signal I (e.g. digital luminance component, colour difference component or composite signal) has the prediction signal P subtracted therefrom in a subtractor 10 to produce an error signal E. This error signal is subjected to a transform 11 such as a discrete cosine transform. The transformed error signal is quantised in a quantizer 12 to produce the coded error signal E which is the DPCM output signal. This signal is subjected to an inverse transform 13 and added to the prediction signal P in adder 14 to produce the locally decoded signal L which constitutes the input to the predictor 15 generating the prediction signal.

Associated with the predictor 15 is a coefficient calculator 16 which calculates the predictor coefficients, based on locally decoded samples of previous fields held in a memory 17.

The following are examples of methods by which the coefficient calculation 16 may calculate a suitable set of predictor coefficients for each block: (i) The adaptive scheme described in Patent Application No.88 00360 may be used, In the mode in which the DPCM predictor is changed, say, once per field while the LMS (least mean squares) predictor from which it is derived may be changed more frequently. The LMS predictor will require also the current error signals, as indicated by a broken line connection from the inverse transform 13 to the coefficient calculator 16 in Fig. I.

(ii) The predictor coefficients may be calculated directly from a phase correlation or cross-correlation function calculated between the current picture area and the corresponding area in the previous field.

(iii) The coefficients may be calculated using an explicit optimization approach based on the autocorrelation function of the picture signal.

These methods of calculating predictor coefficients are not offered as novel in themselves. What is believed to be novel is the use of such a predictor to provide implicit motion compensation in a three-dimensional coding system, particularly a hybrid transform coding system.

Since the methods of coefficient calculation are known in themselves no detailed description of the calculator 16 is given.

Nor is it necessary to describe the predictor 15 which may be conventional. As is well known, a predictor calculates a predicted signal as the sum of a plurality of differently delayed versions of an input signal thereto, multiplied by respective predictor coefficients which, in the case of an adaptive predictor, are not preset but are continuously calculated in dependence upon the signal being coded or encoded. The input signal to the predictor must be the locally decoded signal in the decoder and the locally decoded signal is also used in the coder.

Fig.2 does not differ in principle from Fig.l but the adder 14 and subtractor 10 operate in the transformed plane. Accordingly the transform 11 precedes the subtractor 10 and the adder 14 precedes the inverse transform 13. The predictor signal P is transformed by a transform 18 to a signal P' for application to the subtractor and adder.

Two possible methods by which the decoder may obtain the same predictor coefficients as were used in the coder are envisaged. One method is to transmit the coefficients, calculated in the coder, to the decoder. The other is to ensure that the coder uses in its calculations only information that is also available to the decoder, so that the same calculations (in the absence of transmission errors) may be carried out in the decoder. An advantage of the former method is that it is more robust than the second method in the presence of transmission errors, while an advantage of the second method is that no transmission overhead is incurred. In the second method the samples stored in the memory 17 and used by the coefficient calculator 16 must be taken from the locally decoded signal L,as shown in Figs. 1 and 2. In the first method, the samples could be taken from the input signal I.

Claims

CLAIMS:

1. A coder for a television signal including an adaptive first predictor which calculates a predicted signal and coding means responsive to the input signal and the predicted signal to generate a coded signal, characterized in that the predictor coefficients are provided by calculating means responsive to locally decoded or input samples from a plurality of fields of the television signal, such that the prediction is effective in the temporal, as well as the spatial domain, and thereby effects motion compensation.

2. A coder according to claim 1, wherein the coding means effect differential pulse code modulation.

3. A coder according to claim 1 or 2, wherein the coding means utilize spatial transform coding.

4. A coder according to claim 1, 2 or 3, wherein the calculating means include a supplementary adaptive least means squares predictor operating to adapt its coefficients at a relatively high rate and from which the coefficients of the first predictor are updated at a relatively low rate.

5. A coder according to claim 1, 2 or 3, wherein the calculating means calculate the predictor coefficients from a phase correlation function between the current picture area and the corresponding area in a previous field.

6. A coder according to claim 1, 2 or 3, wherein the calculating means calculate the predictor coefficients from a cross correlation function between the current picture area and the corresponding area in a previous field.

7. A coder according to claim 1, 2 or 3, wherein the calculating means calculate the predictor coefficients in dependence upon the auto correlation function of the television signal.

8. A coder according to any of claims 1 to 7, wherein the predictor coefficients of the first predictor are updated once per field.

9. A decoder complementary to the coder of any preceding claim and including a predictor matching the first predictor of the coder.