초록 ▼

The invention relates to a method and device for estimating motion between images forming a sequence of segmented images or partitions composed of I regions R.sub.i identified by labels. The method comprises a step of initializing motion parameters of each region R.sub.i, an intermediate processing

The invention relates to a method and device for estimating motion between images forming a sequence of segmented images or partitions composed of I regions R.sub.i identified by labels. The method comprises a step of initializing motion parameters of each region R.sub.i, an intermediate processing step of the images on which the estimation of the motion is performed, and an iterative refining step for the definitive determination of said motion parameters in the form of a vector (D.sub.x, D.sub.y) for all the pixels of each of the regions R.sub.i. This refining step, including a minimizing operation of the prediction errors over each object by means of the minimization of an associated function, allows to select the most representative points and reject the others as outliers. Application: very low bitrate coding, multimedia applications.

대표청구항 ▼

[ What is claimed is:] [1.] A method of estimating motion between images forming a sequence P(t-n), P(t-n+1), . . . , P(t-2), P(t-1), P(t), . . . , and available in the form of a sequence S(t-n), S(t-n+1), . . . , S(t-2), S(t-1), S(t), . . . , of segmented images or partitions composed of I regions

[ What is claimed is:] [1.] A method of estimating motion between images forming a sequence P(t-n), P(t-n+1), . . . , P(t-2), P(t-1), P(t), . . . , and available in the form of a sequence S(t-n), S(t-n+1), . . . , S(t-2), S(t-1), S(t), . . . , of segmented images or partitions composed of I regions R.sub.i identified by labels, said method comprising, for supplying for each region R.sub.i an information M.sub.i (t) representative of the motion of the current image P(t) with respect to the previous image P(t-1), the following three operations, performed for each region of said current image:(1) a first step of initializing motion parameters of each region R.sub.i of P(t) as a function of the images P(t-1), P(t) before segmentation and S(t-1), S(t) after segmentation, and of the motion information M.sub.i (t-1) estimated for the previous image P(t-1) in a previous performance of the method;(2) a second step for an intermediate processing of the images on which the estimation of the motion is performed, and a third refining step for the definitive determination of said motion parameters in the form of a vector (Dx, Dy) for all the pixels of each of the regions R.sub.i, in such a way that, for each coordinate point (x,y) of the region, L(x,y,t)=L(x-Dx, y-Dy, t-1), L(.) designating the luminance or a more complex video signal and Dx, Dy being polynomials the degree of which is related to the type of motion of the region;(3) the iterative repetition of said second and third steps of intermediate processing and refinement, until the end of this iterative process as a function of at least a given criterion so as to finally obtain the motion information M.sub.i (t); racterized in that said third refining step includes a minimizing operation of the prediction errors over each object whose motion is estimated, by means of the minimization of an associated function .rho.(x), called objective function, according to the formula min .SIGMA..sub.p .rho.(r.sub.p) where r.sub.p is for each pixel p of each object the prediction error, a prediction error being defined as the difference between original and predicted values, and the motion estimation being based on the assumption that the luminance changes in time are only due to said motion, without consideration of local illumination changes.