Adaptive motion vector field coding

1. A video encoder for motion compensated encoding of video data comprising:

means for generating a first motion vector for a first image segment S K of a current frame to be encoded, based on a reference frame;

motion field encoding means for representing the first motion vector as a first motion coefficient vector c K comprising a first set of absolute motion coefficients c i , such that c i ·f i (x,y) represents an approximation of the first motion vector, f i (x,y) defines a set of basis functions and i is an index that defines correspondences between particular basis functions and absolute motion coefficients c i ;

means for generating a second motion vector for a second image segment S L of the current frame to be subsequently encoded, based on the reference frame;

motion field encoding means for representing the second motion vector as a second motion coefficient vector c L comprising a second set of absolute motion coefficients c j , such that c j ·f j (x,y) represents an approximation of the second motion vector, f j (x,y) defines a set of basis functions and j is an index that defines correspondences between particular basis functions and absolute motion coefficients c j ;

motion coefficient prediction means for predicting a predicted motion coefficient vector p L for the second image segment S L , the predicted motion coefficient vector p L comprising a set of predicted motion coefficients p j , based upon the previously generated first motion coefficient vector c K ;

means for representing each of the absolute motion coefficients c j of the second set of absolute motion coefficients as a sum of a corresponding predicted motion coefficient p j and refinement motion coefficient r j , where each of the refinement motion coefficients r j represents the difference between one of the absolute motion coefficients c j and the corresponding predicted motion coefficient p j ;

means for representing the second motion coefficient vector c L as a vector r L of refinement motion coefficients r j .

2. A video encoder according to claim 1 , wherein each of the motion vectors is represented by an affine motion model.

3. A video encoder according to claim 1 , wherein the absolute motion coefficients, the refinement motion coefficients, and the predicted motion coefficients together with the corresponding basis functions represent an affine motion model.

4. A video encoder according to claim 1 , further comprising:

means for computing a first cost representing an amount of information required to represent segment S L using the absolute motion coefficients c j of the second motion vector c L and an amount of image distortion produced when segment S L is thus represented;

means for computing a second cost representing an amount of information required to represent segment S L using the vector r L of refinement motion coefficients r j and an amount of image distortion produced when segment S L is thus represented;

means for identifying the smaller of the first and second costs;

the encoder being arranged to represent segment S L with whichever of the second motion vector c L and the vector r L of refinement coefficients provides the smaller cost.

5. A video encoder according to claim 1 , further comprising:

means for calculating a matrix E L and a vector y L for the second image segment S L ;

means for calculating a predefined distortion measure for the second segment as a function of E L c L −y L , c L being the second motion vector comprising the second set of absolute motion coefficients c j ; and

means for generating a diagonal matrix R L and a corresponding vector z L 0 from the matrix E L and a vector y L .

6. A video encoder according to claim 5 , further comprising means for obtaining matrix R L by decomposing matrix A L =E L T E L into a lower diagonal matrix R L T and its transpose R L so that A L =R L T R L .

7. A video encoder according to claim 5 , further comprising means for obtaining vector z L 0 by calculating d L =E L T y L and solving the set of equations R L T z L 0 =d L .

8. A video encoder according to claim 5 , further comprising means for calculating an output vector z L =z L 0 −R L p L , where p L is the predicted motion coefficient vector for segment S L .

9. A storage device comprising a video encoder according to claim 1 .

10. A mobile station comprising a video encoder according to claim 1 .

11. A mobile telecommunications network comprising a video encoder according to claim 1 .

12. A computer program product stored on a computer readable medium for encoding video data by a video encoder according to claim 1 .

13. A method for motion compensated encoding of video data comprising:

generating a first motion vector for a first image segment S K of a current frame to be encoded, based on a reference frame;

representing the first motion vector as a first motion coefficient vector c K comprising a first set of absolute motion coefficients c i , such that c i ·f i (x,y) represents an approximation of the first motion vector, f i (x,y) defines a set of basis functions and i is an index that defines correspondences between particular basis functions and absolute motion coefficients c i ;

generating a second motion vector for a second image segment S L of the current frame to be subsequently encoded, based on the reference frame;

representing the second motion vector as a second motion coefficient vector c L comprising a second set of absolute motion coefficients c j , such that c j ·f j (x,y) represents an approximation of the second motion vector, f j (x,y) defines a set of basis functions and j is an index that defines correspondences between particular basis functions and absolute motion coefficients c j ;

predicting a predicted motion coefficient vector p L for the second image segment S L , the predicted motion coefficient vector p L comprising a set of predicted motion coefficients p j , based upon the first motion coefficient vector c K ;

representing each of the absolute motion coefficients c j of the second set of absolute motion coefficients as a sum of a corresponding predicted motion coefficient p j and refinement motion coefficient r j , where each of the refinement motion coefficients r j represents a difference between one of the absolute motion coefficients c j and a corresponding predicted motion coefficient p j ;

representing the second motion coefficient vector c L as a vector r L of refinement motion coefficients r j .

14. A method according to claim 13 , wherein each of the motion vectors is represented by an affine motion model.

15. A method according to claim 13 , wherein the absolute motion coefficients, the refinement motion coefficients, and the predicted motion coefficients together with the corresponding basis functions represent an affine motion model.

16. A method according to claim 13 , further comprising:

computing a first cost representing an amount of information required to represent segment S L using the absolute motion coefficients c j of the second motion vector c L and an amount of image distortion produced when segment S L is thus represented;

computing a second cost representing an amount of information required to represent segment S L using the vector r L of refinement motion coefficients r j and an amount of image distortion produced when segment S L is thus represented;

identifying the smaller of the first and second costs; and

representing segment S L with whichever of the second motion vector c L and the vector r L of refinement coefficients that provides the smaller cost.

17. A method according to claim 13 , further comprising:

calculating a matrix E L and a vector y L for the second image segment S L ;

calculating a predefined distortion measure for the second segment S L as a function of E L c L −y L , c L being the second motion vector comprising the second set of absolute motion coefficients c j ; and

generating a diagonal matrix R L and a corresponding vector z L 0 from the matrix E L and a vector y L .

18. A method according to claim 17 , further comprising obtaining matrix R L by decomposing matrix A L =E L T E L into a lower diagonal matrix R L T and its transpose R L so that A L =R L T R L .

19. A method according to claim 17 , further comprising obtaining vector z L 0 by calculating d L =E L T y L and solving the set of equations R L T z L 0 =d L .

20. A method according to claim 17 , further comprising:

calculating an output vector z L =z L 0 −R L p L , where p L is the predicted motion coefficient.