GB2266639A

GB2266639A - Motion estimation.

Info

Publication number: GB2266639A
Application number: GB9312796A
Authority: GB
Inventors: Michael James Knee
Original assignee: British Broadcasting Corp
Current assignee: British Broadcasting Corp
Priority date: 1989-09-20
Filing date: 1993-06-21
Publication date: 1993-11-03
Anticipated expiration: 2013-06-21
Also published as: GB9312796D0; GB2266639B

Abstract

An apparatus for processing television signals for transmission whereby motion estimation is carried out by calculating a correlation function C(x,y) for motion vectors (x,y) using information from signals representing a previous picture and the current picture. The best sub-integer-pixel motion vector is determined from the calculated values of the correlation function C(x,y), with the calculated values of C(x,y) for non-integer-pixel motion vectors being derived from a comparison of estimated information obtained from the present picture signal and information from the previous picture signal. <IMAGE>

Description

MOTION ESTIMATION This application is divided out of application 8921296.3 which describes another method of motion estimation for use with television signals.

The present invention relates to a method of motion estimation for use with television signals, in particular, to methods capable of providing half-pixel accuracy and using block matching.

In a conventional approach to block matching, a block of samples in the current frame is compared with corresponding previous frame information which has been shifted according to a possible motion vector. The motion vector giving the smallest difference between the current block and the shifted block in the previous frame is selected as the motion estimate.

The invention is defined in the claims appended hereto, to which reference should now be made.

A method in accordance with the invention will now be described in detail, by way of example, with reference to the drawing, which is a block diagram of a motion estimator in accordance with the invention.

In one known method capable of providing half-pixel accuracy, bilinear interpolation is performed on the previous picture to obtain the necessary intermediate pixels for a full search to half pixel accuracy. This constitutes, essentially, a full search to half-pixel accuracy. In a particular implementation in which motion vectors are calculated over a possible range of + 8 pixels in each direction, this requires 32 Motion Estimation Ingetrated Circuits (MEIC) rather than the 8 required for an integer pixel motion estimator. It also requires four versions of the previous picture to be generated, one for each possible pair of fractional values of the motion vector. Thus, as illustrated in the drawing, four half-integer pixel values would be generated by means of bilinear interpolators 20 from the current picture input.

Integer-pixel motion estimators 22 would produce for these values correlation values, the best estimate being selected from the four correlation values by means of a comparison unit 24.

Each version would form the input to an integer-pixel motion estimator consisting off 8 MEICs, via six 1024-pixel buffer stores so 24 such buffer stores would be required. Performing the bilinear interpolation on the three stripes of previous picture information, that is, groups of eight consecutive lines, would also be costly in hardware.

We have appreciated that an alternative would be to generate four versions of the current picture. Thus, for each possible integer motion vector (x,y), vectors (x,y), (x+,y), (x,y+F) and (x+,y+) are tested using current picture signals P(i,j), P(i-,j), P(i,j-) and P(i-,j-) where i and j represent pixel coordinates.

Thus, half-pixel accuracy is obtained by shifting (interpolating) the current block by half-pixel shifts and comparing this shifted block with the previous frame block (shifted by the best integer-pixel motion estimate). There is a slight error in the location of picture material used to test the half-pixel motion vectors, but there is also a substantial saving in hardware. Six buffer stores containing previous-picture information would be sufficient to feed all 32 MEICs and there would only be a comparatively slight increase in the buffer storage required for the input of the current-picture information.

Claims

1. A method of motion estimation for use with television signals, the method comprising calculating a correlation function C(x,y) for sub-integer-pixel motion vectors using information from signals representing a previous picture and the current picture and determining from the calculated values of C(x,y) the best sub-integer-pixel motion vector; the calculated values of C(x,y) for non-integer-pixel motion vectors being derived from a comparison of estimated information obtained from the present picture signal and information from the previous picture signal.

2. A method according to claim 1 in which the correlation function C(x,y) is calculated for half-integer-pixel motion vectors to give the motion vector to half-integer-pixel accuracy.

3. A method according to claim 2 in which information is obtained from the present picture signal by bilinear interpolation to enable non-integer-pixel values of the correlation function C(x,y) to be calculated.

4. A method according to claim 1 substantially as hereinbefore described.

5. Apparatus for carrying out a method of motion estimation for use with television signals according to any of claims 1 to 4.