Abstract
The invention is a method of motion estimation for image compression, firstly decide the initial single body,composed of three points, by utilizing the current graph frame and the block motion vector in previous frame. Subsequently, replace the point with the biggest function value in these three points by a point with relatively small function value to form another single body until there are two points of these three points converges to the same point. A substitution step for the above-mentioned way is to proceed by utilizing downhill single-body search, i.e. by executing steps of reflection, expansion, contraction, or shrinking to find a substitute point. The invention also can utilize a multiple reference-frame method, like the current image frame with respective to the motion vectors in the mentioned plurality of image frames, to decide a single body composed of four points, and then reiteratively replace the point with the biggest function value in these tfour points by a point with relatively small function value to form a single body until there are two points of these four points converges to the same point.