Method of estimation and of hierarchised coding of the motion of image sequences

Claim 1

A method of determining motion of a sequence of images, comprising the steps of: partitioning, at a first level, a current image into macroblocks of 2 p ×2 p pixels and estimating a first motion-vector field associated with said macroblocks at the first level using an iterative and recursive estimation algorithm initialized with motion vectors estimated for a previous image; partitioning, at a second level, each of the macroblocks into quadrants and estimating, for blocks which result therefrom, a second motion-vector field using the same estimation algorithm used for the first level initialized with the vectors which were estimated at the first level; partitioning, at an i th level, i=3 to p, each of the blocks at level i-1 into quadrants and estimating, for blocks which result therefrom, an i th motion-vector field using the same estimation algorithm used for the i-1 level initialized with the vectors which were estimated at the i-1 level, a minimum size of the i th level blocks being no smaller than 2 1 ×2 1 pixels; determining differences in luminance between the pixels of successive images of corresponding blocks at a highest level of partitioning which are displaced by each of the estimated motion vector fields of the p levels of division; summing, for each of said estimated motion vector fields, said differences in luminance of the pixels within the blocks at the highest level of partitioning; and selecting, for each of said blocks at the highest level of partitioning, a final motion vector field, from said motion vector fields, which has a smallest sum of said differences.

Claim 2

A method of determining motion of a sequence of images, comprising the steps of: partitioning, at a first level, a current image into macroblocks of 2 p ×2 p pixels and estimating a first motion-vector field associated with said macroblocks at the first level using an iterative and recursive estimation algorithm initialized with motion vectors estimated for a previous image; partitioning, at a second level, each of the macroblocks into quadrants and estimating, for blocks which result therefrom, a second motion-vector field using the same estimation algorithm used for the first level initialized with the vectors which were estimated at the first level; partitioning, at an i th level, i=3 to p, each of the blocks at level i-1 into quadrants and estimating, for blocks which result therefrom, an i th motion-vector field using the same estimation algorithm used for the i-1 level initialized with the vectors which were estimated at the i-1 level, a minimum size of the i th level blocks being no smaller than 2 1 ×2 1 pixels; determining differences in luminance between the pixels of successive images of corresponding blocks at a highest level of partitioning which are displaced by each of the estimated motion vector fields of the p levels of division; summing, for each of said estimated motion vector fields, said differences in luminance of the pixels within the blocks at the highest level of partitioning; and selecting, for each of said blocks at the highest level of partitioning, a final motion vector field, from said motion vector fields, which has a smallest sum of said differences.

Claim 6

A method according to one of claims 2 and 3, wherein: the correction of the motion vectors is performed by reducing the motion vectors by a quantity dependent on displaced interframe differences in luminance by reducing the motion vectors by a quantity dependent on the sums of differences in luminance between the pixels of the current block and the pixels of the blocks corresponding to the current block and displaced by the motion vector, in at least one of the previous frame and the next frame.

Claim 9

A method according to claims 2 or 3, wherein: the correction of the motion vector is carried out by block matching, over a limited window.

Claim 11

A method according to claims 2 or 3, wherein: when the estimation algorithm used is non-symmetric, the differences in luminance between blocks as well as the corrections are calculated over one of the previous frame and the next frames and are set to one of zero and a same value for the other frame.

Claim 12

A method according to any one of claims 1 to 3, wherein: the final motion-vector field is rendered homogeneous by eliminating from the final motion-vector field, least frequently appearing vectors and by replacing the least frequently appearing vectors with closest vectors, of larger frequency.