US 4,989,087 A · Multi-predictive method to estimate the motion of the points of an electronic picture

Abstract

A method for the estimation of the motion of electronic picture points, in the plane of the picture, by means of fast, convergent and adaptive processing, notably in order to enable subsequent, efficient processing, by sub-sampling, of the picture sequence without harming the psycho-visual quality of the picture reconstructed from the sub-sampled picture. To each electronic picture point, there is assigned a displacement vector, computed according to a WALKER and RAO type of algorithmic process of prediction/correction, the predicted value of motion of the current point being the value of the original displacement vector associated with a point of origin of the preceding picture, so that said current point is the approximate projection of said point of origin along said original displacement vector.

Claims13

Dependent claims

An image signal compaction process of motion estimation for assigning at instant t to each current point comprising a first image, a first value of a motion vector in a plane of said first image, whereby the position of each current point of said first image may be accurately approximated at an instant (t+1) in a later, second image, said compaction process comprising the steps of:

(a) predicting said first value of said motion vector of each current point of said first image by identifying it with a second value of said motion vector associated with a corresponding, antecedent point of a third image at an earlier instant (t-1), said current point being the approximate projection of said antecedent point along said second value of said motion vector, each of said first, second and third images having a characteristic, and
(b) correcting said predicted first value of said motion vector by optimization in accordance with the differences between said characteristics of said first and second images, said predicted first value of said motion vector being taken as a starting hypothesis for said step (b) of correcting.

A process for estimating the motion of a plurality of points comprising first and second images occurring sequentially at instants t and (t+1) respectively, each of said first and second images having a characteristic, said motion estimating process comprising the steps of:

(a) providing for each of said plurality of points a motion vector;
(b) estimating a predicted value of each motion vector from a current point of said first image to project approximately the position of a corresponding point in said second image;
(c) predicting an initial value of said motion vector;
and
(d) correcting said initial value of said motion vector in accordance with an algorithmic prediction/correction process comprising the steps of:
computing in parallel first and second predicted values of each motion vector using a distinct, predicted value of said current point, said predicted value being selected in accordance with at least one spatial prediction of motion, one temporal prediction of motion, or one spatial-temporal prediction in the axis of said motion vector;
and
selecting one of said first and second predicted values of said motion vector according to a decision of optimizing dependent on the difference in the values of said characteristic of said first and second images.

5 more elements

Dependencies are inferred from the claim text.

Multi-predictive method to estimate the motion of the points of an electronic picture

Life of this patent

Multi-predictive method to estimate the motion of the points of an electronic picture

Life of this patent

Abstract

Claims13